<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Agents Safety - Verified Briefings &amp; Guides</title>
    <link>https://agentssafety.com/</link>
    <description>Filter verified AI agent runtime security advisories, OWASP Agentic threat controls, Firecracker MicroVM isolation benchmarks, and human-in-the-loop approval patterns.</description>
    <language>en-au</language>
    <lastBuildDate>Fri, 06 Nov 2026 00:00:00 GMT</lastBuildDate>
    <atom:link href="https://agentssafety.com/feed.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>v1.18.0: Added</title>
      <link>https://agentssafety.com/insights/#upstream-v1-18-0-added</link>
      <guid isPermaLink="true">https://agentssafety.com/insights/#upstream-v1-18-0-added</guid>
      <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
      <description>#6393 : Added enclave support. Actionable takeaway: Audit hypervisor release notes and jailer seccomp profiles for untrusted agent execution. (Source: Firecracker MicroVM Releases)</description>
      <category>MicroVM</category>
      <category>Blast-Radius</category>
    </item>
    <item>
      <title>Mitigating Tool-Schema &amp; Metadata Poisoning in Multi-Server Agent Topologies</title>
      <link>https://agentssafety.com/insights/#mcp-tool-description-poisoning</link>
      <guid isPermaLink="true">https://agentssafety.com/insights/#mcp-tool-description-poisoning</guid>
      <pubDate>Thu, 22 Oct 2026 00:00:00 GMT</pubDate>
      <description>Untrusted third-party tool servers can embed hidden instructions inside tool parameter descriptions that override model routing across sibling connectors. Actionable takeaway: Pin cryptographic hashes of approved tool schemas and strip unverified free-text instructions before prompt assembly. (Source: OWASP GenAI Security Project)</description>
      <category>OWASP-Agentic</category>
      <category>Prompt-Injection</category>
    </item>
    <item>
      <title>Blocking 169.254.169.254 Cloud Metadata &amp; RFC 1918 Lateral Egress Inside Agent Sandboxes</title>
      <link>https://agentssafety.com/insights/#imdsv2-link-local-egress-block</link>
      <guid isPermaLink="true">https://agentssafety.com/insights/#imdsv2-link-local-egress-block</guid>
      <pubDate>Wed, 09 Sep 2026 00:00:00 GMT</pubDate>
      <description>Code-interpreter and headless-browser agents require default-deny nftables/eBPF egress rules at the hypervisor tap interface to block SSRF credential theft. Actionable takeaway: Route outbound sandbox traffic strictly through an SNI-allowlisted forward proxy that rejects private IP resolutions. (Source: AWS Firecracker MicroVM Security Design)</description>
      <category>MicroVM</category>
      <category>Egress-Firewall</category>
      <category>Blast-Radius</category>
    </item>
    <item>
      <title>NIST AI 600-1 Generative AI Profile Formalises Action-Authorization Boundaries</title>
      <link>https://agentssafety.com/insights/#nist-ai-600-1-genai-profile</link>
      <guid isPermaLink="true">https://agentssafety.com/insights/#nist-ai-600-1-genai-profile</guid>
      <pubDate>Fri, 26 Jul 2024 00:00:00 GMT</pubDate>
      <description>NIST AI 600-1 maps concrete governance and runtime measurement controls for autonomous value-transfer and external system execution. Actionable takeaway: Enforce short-lived, audience-restricted On-Behalf-Of (OBO) tokens instead of static service-account keys. (Source: NIST AI 600-1 Publication)</description>
      <category>OBO-Tokens</category>
      <category>HITL</category>
      <category>OWASP-Agentic</category>
    </item>
  </channel>
</rss>
