<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>irzubal on irzubal · AI security incident analysis</title><link>https://irzubal.com/en/</link><description>Recent content in irzubal on irzubal · AI security incident analysis</description><generator>Hugo</generator><language>en-US</language><lastBuildDate>Sun, 27 Sep 2026 00:00:00 +0300</lastBuildDate><atom:link href="https://irzubal.com/en/index.xml" rel="self" type="application/rss+xml"/><item><title>EchoLeak: Leaking data from Microsoft 365 Copilot with a single email</title><link>https://irzubal.com/en/incidents/echoleak-m365-copilot/</link><pubDate>Sun, 27 Sep 2026 00:00:00 +0300</pubDate><guid>https://irzubal.com/en/incidents/echoleak-m365-copilot/</guid><description>Aim Labs showed that M365 Copilot could follow hidden instructions in an incoming email and carry internal data out of the organisation without the user clicking anything. Microsoft fixed it server-side.</description></item><item><title>About</title><link>https://irzubal.com/en/about/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://irzubal.com/en/about/</guid><description>&lt;h2 id="why-this-blog-exists"&gt;Why this blog exists&lt;/h2&gt;&#10;&lt;p&gt;In AI security, the best textbook is real incidents. Here I take publicly disclosed AI security incidents and read them from both sides: what the attacker saw, what the defence missed, and what I would have done. Every analysis follows the same six steps and is mapped to the OWASP LLM Top 10 and MITRE ATLAS, so over time the posts become a searchable incident archive.&lt;/p&gt;</description></item><item><title>Index</title><link>https://irzubal.com/en/tags/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://irzubal.com/en/tags/</guid><description/></item></channel></rss>