Facio Blog
Payload-powered product notes, security writing, HITL patterns, and operational guidance from the Facio runtime: long sessions, Placet approvals, audit trails, memory, providers, channels, tools, and Docker-first operations.
Showing 21-25 of 44 articles in Security.
Security
A malicious MCP server doesn't need to be called to compromise your agent. Hidden instructions in tool descriptions — invisible to humans, fully visible to LLMs — execute as soon as the schema enters the model context. CVE-2025-6514 infected 437,000 installs. The rug pull is the worst part.
Security
Microsoft added "AI" as a fifth pillar to its Zero Trust Workshop with 700 controls. Anthropic published a three-tier framework the same week. CISA's ATF from CSA, the NSA's MCP guidance, and the OWASP Top 10 all point to the same answer. The convergence is the story.
Security
Traditional DLP inspects the perimeter. AI agents operate inside it — reading customer records, summarizing financial PDFs, calling APIs. 78% of employees already use unsanctioned AI tools. Your content inspection was designed for a world where users paste data into form fields. Now the form fields think.
Security
Meta's hallucinated permission exposure, the Mercor LiteLLM supply chain breach, AI-generated polymorphic malware, coordinated AI-DDoS-API campaigns, AI agent control failures, and the model leak fallout. Each incident dissected with full MITRE ATT&CK mapping.
Security
Traditional monitoring can keep 99.99% availability while the agent's decisions degrade quality across the workflow. Decision integrity is the new observability surface — and OpenTelemetry's GenAI semantic conventions plus the new MCP tracing layer are finally making it tractable. Here's how to build for it.