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 16-20 of 44 articles in Security.
Security
You get paged at 3 AM. The agent is calling APIs at 10x normal rate, your SIEM is silent, your audit trail has the first 200 events but not the last 8,000, and the compliance officer needs an incident timeline by 9 AM. The agent runtimes that survive this are the ones built for it.
Security
Static system prompts are advisory, not architectural. LLM guardrails operate at the same abstraction as the attack. The only deterministic defense is policy at the execution layer — where tool calls become real actions against real resources, with sub-millisecond evaluation.
Security
Eleven risk categories. Each one a real production incident. Each one requiring a different defensive control. The OWASP Agentic AI Threats and Mitigations Guide v1.0 went stable in March 2026. The risk-to-defense mapping is what your security team actually needs.
Security
ModelScope's MS-Agent framework shipped a critical command injection vulnerability (CVSS 9.8) that turns any document, ticket, or email into a remote shell for attackers. No patch. Three months after disclosure. The procurement agent that approved $5M in fraudulent orders wasn't using a vulnerable framework — it was using any framework, the same way every agent does.
Security
Traditional SDLC assumes humans write code, humans review code, and security checks happen at fixed phases. AI breaks all three assumptions. Code, dependencies, CI workflows, and infrastructure now arrive together — pre-generated, pre-approved, and pre-merged before any human reviews them.