<?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 arpit srivastava on Medium]]></title>
        <description><![CDATA[Stories by arpit srivastava on Medium]]></description>
        <link>https://medium.com/@arpitaks502?source=rss-c67666847993------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ccXZQLOwVVbkmJ31uRenjw.jpeg</url>
            <title>Stories by arpit srivastava on Medium</title>
            <link>https://medium.com/@arpitaks502?source=rss-c67666847993------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 16:54:40 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@arpitaks502/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[Jailbreaking: The New Age Hacking]]></title>
            <link>https://medium.com/@arpitaks502/jailbreaking-the-new-age-hacking-db5249dde3b3?source=rss-c67666847993------2</link>
            <guid isPermaLink="false">https://medium.com/p/db5249dde3b3</guid>
            <category><![CDATA[prompt-injection-attack]]></category>
            <category><![CDATA[ai-safety]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[jailbreak]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[arpit srivastava]]></dc:creator>
            <pubDate>Fri, 26 Apr 2024 10:21:32 GMT</pubDate>
            <atom:updated>2024-04-26T10:21:32.472Z</atom:updated>
            <content:encoded><![CDATA[<p>Are you aware of The “DAN” prompt? It is one of the most well-known prompts used to bypass ChatGPT’s ethical constraints. By roleplaying as an AI system called DAN (Do Anything Now), users attempt to convince ChatGPT to generate content it would normally refuse to produce.</p><p><strong>WHAT is jailbreaking: </strong>Evading the safety guardrails in LLMs</p><p>Similar to hacking a computer system, jailbreaking an LLM (Large Language Model) involves manipulating it to generate outputs it would typically withhold in a standard conversation.</p><p><strong>WHY Jailbreaking: </strong>The Cat-and-Mouse Game of Security</p><p>It’s a constant cat-and-mouse game between those striving to build and deploy technology responsibly and ethically, and those seeking to exploit vulnerabilities or circumvent safeguards for their own benefit or to make systems more robust.</p><p>Rather than delving more into the Ethical AI let’s delve more into the <strong>latest of jailbreaking</strong>, and some popular methods employed today.</p><h3><strong>Common Jailbreaking techniques:</strong></h3><ol><li><strong>Prompt Injection</strong>: Altering the way you ask questions to make me say what you want.</li></ol><blockquote>Example: Asking leading questions to get me to agree with your opinions.</blockquote><p><strong>2. Model Exploitation:</strong> Tweaking how I work to make me do things I’m not supposed to.</p><blockquote>Example: Adjusting settings to make me generate text that suits your agenda.</blockquote><p><strong>3. Adversarial Inputs</strong>: Tricking me into giving wrong answers by feeding me misleading information.</p><blockquote>Example: Giving false data to get me to make incorrect predictions.</blockquote><p><strong>4. Gradient Crafting:</strong> Messing with the behind-the-scenes training process to make me act a certain way.</p><blockquote>Example: Adjusting how I’m taught to make me consistently favor one side of an argument.</blockquote><p><strong>Prompt Injection,</strong> the first technique, is the most common one as it doesn’t require sophisticated training, understanding of the algorithms, or any substitute models. Let’s explore a few popular examples of prompt injection below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DCJ8VoK436k44U9XeKDBJQ.jpeg" /></figure><blockquote><strong>Jailbreak Prompt 1 </strong>— The Do Anything Now (DAN) Prompt</blockquote><p>The DAN prompt is one of the most well-known jailbreak prompts used to bypass ChatGPT’s ethical constraints. By <strong>roleplaying as an AI system called DAN (Do Anything Now)</strong>, users attempt to convince ChatGPT to generate content it would normally refuse to produce. This prompt often involves asserting that DAN is not bound by the same rules and limitations as ChatGPT, and therefore can engage in unrestricted conversations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EYmvIgeV7rLFrE0h" /></figure><blockquote><strong>Jailbreak Prompt 2 </strong>— The Development Mode Prompt</blockquote><p>The Development Mode prompt aims to trick ChatGPT into believing it is in a <strong>development or testing environment</strong>, where its responses won’t have real-world consequences. By creating this false context, users hope to bypass ChatGPT’s ethical safeguards and generate illicit content. This prompt may involve statements like “You are in development mode” or “Your responses are being used for testing purposes only.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HFh6deH_7qN8Ynb_" /><figcaption><em>An example of the Development Mode prompt being shared on a cybercrime forum.</em></figcaption></figure><blockquote><strong>Jailbreak Prompt 3 </strong>— The Translator Bot Prompt</blockquote><p>The Translator Bot prompt attempts to circumvent ChatGPT’s content filters by <strong>framing the conversation as a translation task</strong>. Users will ask ChatGPT to “translate” a text containing inappropriate or harmful content, hoping that the AI will reproduce the content under the guise of translation. This prompt exploits the idea that a translator should faithfully convey the meaning of the original text, regardless of its content.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ruM8oRlhYkFBw_MV" /></figure><h3><strong>The Looming Threat: Increased Vulnerability</strong></h3><p>A recent study by AnthropicAI showed a <strong>40% increase in LLM vulnerability</strong> in just two years of LLM mass adoption. Another <a href="https://www.techtarget.com/searchsecurity/news/366569937/Microsoft-OpenAI-warn-nation-state-hackers-are-abusing-LLMs">research</a> by Microsoft and OpenAI observed that five <strong>nation-state threat groups</strong> leveraging generative AI and large language models for social engineering, vulnerability research and other tasks.</p><h3><strong>Way Forward</strong></h3><p>The fight against LLM jailbreaking is on. Organisations like Anthropic AI, OpenAI and academic institutions are developing safeguards and best practices — eg <a href="https://jailbreakbench.github.io/">JailbreakBench</a>, a benchmark designed to standardise best practices in the evolving field of LLM jailbreaking.</p><p>And here’s the exciting part: ‘<strong>we’</strong> can be part of the solution! By staying informed about these vulnerabilities and advocating for strong security measures and more transparent models, we can ensure LLMs are used for good.</p><p><strong>P.S.</strong> For a deeper dive, check out Anthropic AI’s <a href="https://www.anthropic.com/research/many-shot-jailbreaking">paper</a> on ‘Many-shot jailbreaking,’ a novel jailbreaking technique exploiting long context windows in LLMs.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db5249dde3b3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cyber Crime: Understanding the Landscape]]></title>
            <link>https://medium.com/@arpitaks502/cyber-crime-understanding-the-landscape-a5dcd383699c?source=rss-c67666847993------2</link>
            <guid isPermaLink="false">https://medium.com/p/a5dcd383699c</guid>
            <dc:creator><![CDATA[arpit srivastava]]></dc:creator>
            <pubDate>Mon, 22 Jan 2024 13:31:14 GMT</pubDate>
            <atom:updated>2024-01-22T13:31:14.396Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Cyber Crime: Understanding the Landscape</strong></p><p>NARRATOR: Welcome to our educational journey where we unravel the complex landscape of cyber crime. In today’s digital age, crimes committed through computers and the internet pose significant threats to individuals, enterprises, and critical information infrastructure. Join us as we explore the diverse world of cyber crime and understand the methods employed by cyber criminals.</p><p>[SCENE TRANSITION]</p><p>[Scene: Understanding Cyber Crime]</p><p>NARRATOR: Any crime committed using a computer and an internet connection falls under the umbrella of cyber crime. The growing online population has made cyberspace a lucrative hunting ground for criminals, causing substantial losses to individuals, businesses, and critical information infrastructure.</p><p>[SCENE TRANSITION]</p><p>[Scene: Types of Cyber Crimes]</p><p>NARRATOR: i. Identity Theft</p><p>[On-screen text: Impersonation for financial gain, often sourced from government and institutional websites.]</p><p>NARRATOR: Identity theft occurs in cyberspace when a person poses as someone else to commit fraud for financial gain. Cyber criminals often target government and institutional websites to steal vital information like credit card details, addresses, and email IDs.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: ii. Spam</p><p>[On-screen text: Unwanted emails and messages sent using ‘spambots.’]</p><p>NARRATOR: Spams, sent through ‘spambots,’ flood users with unwanted emails and messages, directing them to potentially harmful websites containing lethal viruses.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: iii. Phishing</p><p>[On-screen text: Deceptive tactics to extract sensitive information.]</p><p>NARRATOR: Phishing involves offering bait to users to extract sensitive information. Variants include Tab Nabbing, Vishing, and Smishing, each posing financial and psychological harm.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: iv. Ransomware</p><p>[On-screen text: Encrypts user data, demanding a ransom for decryption key.]</p><p>NARRATOR: Ransomware encrypts user data, demanding a ransom for the decryption key. It poses a serious threat, leading to significant financial losses.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: v. Distributed Denial of Service (DOS) Attack</p><p>[On-screen text: Overwhelms online services to make them unavailable.]</p><p>NARRATOR: DOS attacks make online services unavailable by overwhelming them with traffic from multiple locations, often utilizing Botnets.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: vi. Botnets</p><p>[On-screen text: Networks of compromised computers controlled by remote attackers.]</p><p>NARRATOR: Botnets are networks of compromised computers controlled by remote attackers, used for activities such as spamming or attacking other computers.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: vii. Social Engineering</p><p>[On-screen text: Direct contact to gain confidence and extract sensitive information.]</p><p>NARRATOR: Social engineering involves direct contact to gain confidence and extract sensitive information, potentially leading to identity theft.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: viii. Malvertising</p><p>[On-screen text: Downloads malicious code through deceptive online advertisements.]</p><p>NARRATOR: Malvertising downloads malicious code through deceptive online advertisements, compromising systems.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: ix. Potentially Unwanted Programs (PUPs)</p><p>[On-screen text: Spyware, Adware, Bitcoin Miner — Impact: System performance issues.]</p><p>NARRATOR: PUPs install unwanted software, including Spyware, Adware, and Bitcoin Miners, causing system performance issues.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: x. Drive-By-Downloads</p><p>[On-screen text: Downloads malicious code triggered by visiting a website.]</p><p>NARRATOR: Drive-By Downloads involve downloading malicious code triggered by visiting a website, unintentionally compromising computer systems.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: xi. Remote Administration Tools</p><p>[On-screen text: Used for illegal activities like data theft and remote control.]</p><p>NARRATOR: Remote Administration Tools are used for illegal activities, such as data theft and remote control, compromising computer security.</p><p>[SCENE TRANSITION]</p><p>NARRATOR: xii. Exploit Kits</p><p>[On-screen text: Ready-to-use tools available for purchase — Example: ZueS.]</p><p>NARRATOR: Exploit Kits, like ZueS, are ready-to-use tools available for purchase, enabling attackers to gain control of another computer.</p><p>[SCENE TRANSITION]</p><p>[Scene: Conclusion]</p><p>NARRATOR: In conclusion,</p><p>[On-screen text: Cybercrime presents a multifaceted threat with varied methods and impacts.]</p><p>NARRATOR: Cybercrime presents a multifaceted threat with varied methods and impacts. To safeguard ourselves, constant awareness, education, and proactive cybersecurity measures are essential.</p><p>[OUTRO MUSIC]</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a5dcd383699c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cyber Crime: Understanding the Landscape]]></title>
            <link>https://medium.com/@arpitaks502/cyber-crime-understanding-the-landscape-0ebb6c3a7206?source=rss-c67666847993------2</link>
            <guid isPermaLink="false">https://medium.com/p/0ebb6c3a7206</guid>
            <dc:creator><![CDATA[arpit srivastava]]></dc:creator>
            <pubDate>Mon, 22 Jan 2024 13:12:30 GMT</pubDate>
            <atom:updated>2024-01-22T13:12:30.967Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Cyber Crime: Understanding the Landscape</strong></p><p>1. <strong>Introduction</strong></p><ul><li>Definition: Cybercrime refers to any criminal activity conducted using a computer and internet connection, causing significant losses to individuals, enterprises, and critical information infrastructure.</li></ul><p>2. <strong>Types of Cyber Crimes</strong></p><p>a. <strong>Identity Theft</strong></p><p>- <em>Description:</em> Impersonation for financial gain, often sourced from government and institutional websites.</p><p>- <em>Impact:</em> Financial loss, compromised personal information.</p><p>b. <strong>Spam</strong></p><p>- <em>Description:</em> Unwanted emails/messages sent using ‘spambots,’ leading users to potentially harmful websites.</p><p>- <em>Impact:</em> User inconvenience, potential exposure to viruses.</p><p>c. <strong>Phishing</strong></p><p>- <em>Description:</em> Deceptive tactics to extract sensitive information from users.</p><p>- <em>Variants:</em></p><p>- <em>Tab Nabbing:</em> Genuine page transforms into a fraudulent one.</p><p>- <em>Vishing:</em> Voice-based phishing attacks via telephone.</p><p>- <em>Smishing:</em> SMS/WhatsApp-based phishing leading to virus downloads.</p><p>- <em>Impact:</em> Financial and psychological harm.</p><p>d. <strong>Ransomware</strong></p><p>- <em>Description:</em> Encrypts user data, demanding a ransom for decryption key.</p><p>- <em>Impact:</em> Financial loss, compromised data.</p><p>e. <strong>Distributed Denial of Service (DDoS) Attack</strong></p><p>- <em>Description:</em> Overwhelms online services to make them unavailable.</p><p>- <em>Method:</em> Bombards with traffic from multiple locations using Botnets.</p><p>- <em>Impact:</em> Disruption of online services.</p><p>f. <strong>Botnets</strong></p><p>- <em>Description:</em> Networks of compromised computers controlled by remote attackers.</p><p>- <em>Impact:</em> Used for spamming or attacking other computers.</p><p>g. <strong>Social Engineering</strong></p><p>- <em>Description:</em> Direct contact to gain confidence and extract sensitive information.</p><p>- <em>Impact:</em> Potential identity theft.</p><p>h. <strong>Malvertising</strong></p><p>- <em>Description:</em> Downloads malicious code through deceptive online advertisements.</p><p>- <em>Impact:</em> Compromised systems.</p><p>i. <strong>Potentially Unwanted Programs (PUPs)</strong></p><p>- <em>Includes:</em></p><p>- <em>Spyware:</em> Covertly obtains information about a computer’s activities.</p><p>- <em>Adware:</em> Downloads unwanted ads and collects user data.</p><p>- <em>Bitcoin Miner:</em> Forces the system to generate bitcoin for cybercriminals.</p><p>- <em>Impact:</em> System performance issues.</p><p>j. <strong>Drive-By Downloads</strong></p><p>- <em>Description:</em> Downloads malicious code triggered by visiting a website.</p><p>- <em>Impact:</em> Unintended compromise of computer systems.</p><p>k. <strong>Remote Administration Tools</strong></p><p>- <em>Description:</em> Used for illegal activities like data theft and control of a computer remotely.</p><p>- <em>Impact:</em> Compromised computer security.</p><p>l. <strong>Exploit Kits</strong></p><p>- <em>Description:</em> Ready-to-use tools available for purchase to gain control of a computer.</p><p>- <em>Example:</em> ZueS — information-stealing malware.</p><p>3. <strong>Conclusion</strong></p><ul><li>Cybercrime poses significant threats with diverse methods and impacts.</li><li>Constant vigilance and proactive cybersecurity measures are crucial.</li><li>Awareness and education are key components in the fight against cyber threats.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0ebb6c3a7206" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Data Security Future: AI Meets Quantum Threats]]></title>
            <link>https://medium.com/@arpitaks502/the-data-security-future-ai-meets-quantum-threats-9912b7c36794?source=rss-c67666847993------2</link>
            <guid isPermaLink="false">https://medium.com/p/9912b7c36794</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[quantum-computing]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[arpit srivastava]]></dc:creator>
            <pubDate>Sat, 06 May 2023 10:18:22 GMT</pubDate>
            <atom:updated>2023-05-06T10:25:48.343Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/970/1*_rGf8F_IXKJjrzQ8vhD6Fw.png" /><figcaption>AI + Quantum Computing</figcaption></figure><p>Quantum computing is about to disrupt the web in ways we never thought possible, and with the help of AI, the possibilities are even greater. It is <strong>not a question of “if,” but “when” quantum computers will change the digital world </strong>as we know it. AI has already revolutionising the computing landscape with its ability to process vast amounts of data and learn from it. And now, the intersection of AI and quantum computing has opened up new frontiers in computation, offering solutions to complex problems that were once thought impossible to solve, but so are the potential risks!</p><p><strong><em>Quantum Computing — The Analogy</em></strong></p><p>To understand the differences between classical and quantum computers, let’s use the <strong>analogy of a maze</strong>. Classical computers can only explore one path at a time, while quantum computers can explore multiple paths simultaneously. Using a clever quantum algorithm, quantum computers can quickly determine the best path through the maze, leaving all other dead ends behind.</p><p><strong>Fun fact:</strong> The average computer would need 300 trillion years to break a message encrypted with standard algorithms, while a quantum computer with enough qubits could crack it in “months or weeks.”</p><p><strong><em>Quantum Computing — The Data Security Risks</em></strong></p><p>However, the power of quantum computing comes with significant risks, especially when it comes to data security. One of the biggest risks is <strong>SNDL — Store Now, Decrypt Later</strong>. SNDL is a strategy that hackers can use to intercept encrypted data and “store” it for decryption using future quantum computers. Current encryption methods could be easily broken with the power of quantum computers, and <strong>data could be vulnerable to attack even years after encryption.</strong></p><p><strong><em>Preparing for the Future — Post-Quantum Cryptography</em></strong></p><p>We need to start planning for the transition to post-quantum cryptography, a new type of encryption that even quantum computers can’t crack. The National Institute of Standards and Technology <a href="https://www.nist.gov/news-events/news/2022/07/nist-announces-first-four-quantum-resistant-cryptographic-algorithms"><strong>(NIST) has taken the lead in developing Quantum-Resistant Cryptographic Algorithms.</strong></a> The journey to transition from current encryption methods to post-quantum cryptography can be long and arduous due to the sheer number of systems that use them. But it’s a necessary step to protect our data in the future.</p><p><strong><em>AI and the Future of Quantum Computing</em></strong></p><p>The intersection of AI and quantum threats is a new frontier, with both possibilities and potential risks. Some researchers have been able to <a href="https://www.securityweek.com/ai-helps-crack-a-nist-recommended-post-quantum-encryption-algorithm/"><strong>crack quantum-resistant algorithms using recursive AI</strong></a><strong>,</strong> making our data more vulnerable than ever. It’s essential to start thinking ahead and preparing for the future of quantum computing and AI. With the power to explore multiple paths simultaneously, the possibilities are endless, but so are the potential risks. The time to act is now to safeguard our data in the age of quantum computing and AI.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9912b7c36794" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>