<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Kinwilder · AI &amp; agents</title><link>https://kinwilder.com/</link><description>Cybersecurity, decoded. AI-assisted analysis with primary sources.</description><language>en</language><atom:link href="https://kinwilder.com/feeds/ai.xml" rel="self" type="application/rss+xml"/><lastBuildDate>Sat, 24 Oct 2026 16:13:12 GMT</lastBuildDate><item><title>When a security verdict needs a third value.</title><link>https://kinwilder.com/read/pyrit-the-security-verdict-that-needs-three-values</link><guid isPermaLink="true">https://kinwilder.com/read/pyrit-the-security-verdict-that-needs-three-values</guid><pubDate>Sat, 24 Oct 2026 16:13:12 GMT</pubDate><description>A five-line Microsoft PyRIT helper preserves “undetermined.” The hard part is keeping that meaning intact all the way to a release decision.</description><category>AI security</category></item><item><title>Your AI found a bug. Where is the witness?</title><link>https://kinwilder.com/read/ai-bug-hunting-needs-a-witness</link><guid isPermaLink="true">https://kinwilder.com/read/ai-bug-hunting-needs-a-witness</guid><pubDate>Sat, 24 Oct 2026 05:00:00 GMT</pubDate><description>Google’s PageBreak puts the spotlight on a less glamorous breakthrough: a finding has to survive a test outside the model.</description><category>AI security</category></item><item><title>The prompt is not the permission slip.</title><link>https://kinwilder.com/read/prompt-injection-is-an-authority-problem</link><guid isPermaLink="true">https://kinwilder.com/read/prompt-injection-is-an-authority-problem</guid><pubDate>Sat, 24 Oct 2026 05:00:00 GMT</pubDate><description>Why an agent reading a webpage should not inherit that page’s ambitions—and how to put the boundary in code.</description><category>AI security</category></item></channel></rss>
