<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>CyberStrikeAI on Martin's Blog</title><link>https://mac.sploit.dk/tags/cyberstrikeai/</link><description>Recent content in CyberStrikeAI on Martin's Blog</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sat, 21 Nov 2026 10:00:00 +0100</lastBuildDate><atom:link href="https://mac.sploit.dk/tags/cyberstrikeai/index.xml" rel="self" type="application/rss+xml"/><item><title>Five AI Pentest Frameworks, and Where the Walls Actually Are</title><link>https://mac.sploit.dk/blog/ai-pentest-frameworks-where-the-walls-are/</link><pubDate>Sat, 21 Nov 2026 10:00:00 +0100</pubDate><guid>https://mac.sploit.dk/blog/ai-pentest-frameworks-where-the-walls-are/</guid><description>&lt;p&gt;Autonomous AI pentest frameworks have stopped being a research curiosity. ARTEX forks are showing up in incident reports, Cairn panels are exposed on the internet, and CyberStrikeAI made the news for running 221 tool calls without anyone approving them.&lt;/p&gt;&#10;&lt;p&gt;I wanted to know what these things actually are underneath the README claims. So I read the source of five of them: HexStrike AI, PentAGI, Cairn, ARTEX-ymh and CyberStrikeAI. I looked at three questions. Who plans the work? What executes it? And when a framework says it limits targets or actions, where does that limit live?&lt;/p&gt;&#10;&lt;p&gt;The answer to the last question is the one that matters. Where scope exists at all, it is mostly prompt text or opt-in configuration. Only two of the five ship any default target restriction in code, and both are narrow. None of them requires a human to approve anything out of the box.&lt;/p&gt;</description></item></channel></rss>