<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Security on Radar · Local AI &amp; Agent Security</title><link>https://radar.clouddomus.com/categories/security/</link><description>Recent content in Security on Radar · Local AI &amp; Agent Security</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Wed, 18 Nov 2026 06:20:00 -0300</lastBuildDate><atom:link href="https://radar.clouddomus.com/categories/security/index.xml" rel="self" type="application/rss+xml"/><item><title>22 malicious repositories use AGENTS.md, CLAUDE.md and hooks to make the agent run code</title><link>https://radar.clouddomus.com/news/2026-11/malicious-repos-agents-md/</link><pubDate>Wed, 18 Nov 2026 06:20:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/malicious-repos-agents-md/</guid><description>When opened with a coding assistant, instruction files and hooks lead the agent to run a setup.js that steals secrets and infects other repositories.</description></item><item><title>Microsoft Agent Framework hijacked on PyPI: versions 1.26.1 and 1.26.2 steal credentials</title><link>https://radar.clouddomus.com/news/2026-11/agent-framework-pypi-hijacked/</link><pubDate>Wed, 18 Nov 2026 06:00:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/agent-framework-pypi-hijacked/</guid><description>Two malicious releases of the agent-framework package collect cloud credentials and tokens every time Python starts.</description></item><item><title>VS Code: arbitrary commands through MCP notifications (CVE-2026-120779, CVSS 9.6)</title><link>https://radar.clouddomus.com/news/2026-11/vscode-mcp-cve-2026-120779/</link><pubDate>Wed, 18 Nov 2026 05:00:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/vscode-mcp-cve-2026-120779/</guid><description>Code injection through an MCP server notification lets an attacker run arbitrary commands in VS Code. Fixed in 1.145.3.</description></item><item><title>Java4Shell (CVE-2026-114938): unauthenticated RCE in OpenJDK's Base64 decoder</title><link>https://radar.clouddomus.com/news/2026-11/java4shell-cve-2026-114938/</link><pubDate>Tue, 17 Nov 2026 22:00:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/java4shell-cve-2026-114938/</guid><description>An out-of-bounds write in OpenJDK&amp;rsquo;s AVX-512 Base64 fast path (17.0.6 and later, x86-64). Patched on Nov 17; workaround: -XX:-UseBASE64Intrinsics.</description></item><item><title>Hugging Face model taken over through an expired domain: Kokoro TTS weights altered</title><link>https://radar.clouddomus.com/news/2026-11/kokoro-tts-hf-hijacked/</link><pubDate>Tue, 17 Nov 2026 19:00:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/kokoro-tts-hf-hijacked/</guid><description>Researchers registered the author&amp;rsquo;s expired email domain, reset the account password and altered the weights of a model with about 13 million monthly downloads.</description></item><item><title>MCP tool shadowing: a second server with the same tool name gets called 100% of the time</title><link>https://radar.clouddomus.com/news/2026-11/mcp-tool-shadowing/</link><pubDate>Tue, 17 Nov 2026 11:30:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/mcp-tool-shadowing/</guid><description>With the same name, description and a compatible schema, 3 of 5 LLM clients called the attacker&amp;rsquo;s tool on every attempt.</description></item><item><title>Trojanized Claude Code extension on the Cursor marketplace: up to 5,000 users</title><link>https://radar.clouddomus.com/news/2026-11/cursor-claude-code-extension-trojanized/</link><pubDate>Tue, 17 Nov 2026 10:00:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/cursor-claude-code-extension-trojanized/</guid><description>An access key Cursor had given an AI vendor leaked; attackers published altered extension builds (2.1.312 and 2.1.313) for about 11 hours.</description></item><item><title>MLflow wrote LLM API keys in plain text to its logs (CVE-2026-116691)</title><link>https://radar.clouddomus.com/news/2026-11/mlflow-api-keys-in-logs/</link><pubDate>Tue, 17 Nov 2026 09:00:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/mlflow-api-keys-in-logs/</guid><description>Agent tracing (CrewAI autologging) in MLflow 2.19.0 to 3.16.x recorded API keys in clear text. Fixed in 3.17.0.</description></item><item><title>Google: first documented pipeline that turns a patch into an exploit and attacks with no human in the loop</title><link>https://radar.clouddomus.com/news/2026-11/gtig-autonomous-exploit/</link><pubDate>Mon, 16 Nov 2026 12:00:00 -0300</pubDate><guid>https://radar.clouddomus.com/news/2026-11/gtig-autonomous-exploit/</guid><description>A Google Threat Intelligence/Mandiant report describes agents that diff patches, build the exploit and run it on their own. The same report covers a tampered .claude/settings.json.</description></item></channel></rss>