<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Research on Security Sonar</title><link>https://securitysonar.com/research/</link><description>Recent content in Research on Security Sonar</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sun, 23 Aug 2026 16:00:00 +0000</lastBuildDate><atom:link href="https://securitysonar.com/research/index.xml" rel="self" type="application/rss+xml"/><item><title>The Real AI Attack Surface Isn't the Model — It's the Harness</title><link>https://securitysonar.com/research/ai-harness-is-the-real-attack-surface/</link><pubDate>Sun, 23 Aug 2026 16:00:00 +0000</pubDate><guid>https://securitysonar.com/research/ai-harness-is-the-real-attack-surface/</guid><description>Three independent 2026 research efforts show agents getting compromised with no model jailbreak required — only flaws in the orchestration, tool-use, and guardrail code wrapped around the model. That wrapper is the harness, and it&amp;rsquo;s where the real attack surface lives.</description></item></channel></rss>