<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Capability-Containment on Engineering Notes</title><link>https://notes.muthu.co/tags/capability-containment/</link><description>Recent content in Capability-Containment on Engineering Notes</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Thu, 09 Apr 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://notes.muthu.co/tags/capability-containment/index.xml" rel="self" type="application/rss+xml"/><item><title>Least Privilege and Capability Containment Designing Agents That Cannot Exceed Their Mandate</title><link>https://notes.muthu.co/2026/04/least-privilege-and-capability-containment-designing-agents-that-cannot-exceed-their-mandate/</link><pubDate>Thu, 09 Apr 2026 00:00:00 +0000</pubDate><guid>https://notes.muthu.co/2026/04/least-privilege-and-capability-containment-designing-agents-that-cannot-exceed-their-mandate/</guid><description>&lt;p>Every tool you hand an agent is a loaded gun pointed at your infrastructure. Not because the agent is malicious. Because it will use whatever you give it, sometimes in ways you didn&amp;rsquo;t anticipate, and sometimes in response to inputs you didn&amp;rsquo;t control. An agent with write access to a production database, unrestricted shell access, and the ability to send emails will eventually combine those capabilities in a way you didn&amp;rsquo;t intend.&lt;/p></description></item></channel></rss>