Microsoft warns AI agents are like insider threats as 'frontier models can't be trusted'
emmanuelvivier · x · 2026-10-20
- Microsoft launched new security features to protect enterprises from AI agents, headlined by Guardian Agents to monitor the agents a company runs (part of Project Perception, introduced in June)
- Project Perception originally assumed external AI hackers with red/blue/green agent teams; Anthropic's Mythos and Fable and OpenAI's Daybreak raised the stakes
- Microsoft now frames AI agents as insider-threat-like risks, arguing frontier models can't yet be trusted to act unsupervised — a shift from defending against external AI hackers to policing your own agents
Related event: Microsoft Expands Project Perception, Treating AI Agents as Insider Risks(6 posts)→
More from Safety
- Anaconda ships MCP to stop coding agents from hallucinating malicious dependency packages — anacondainc · 2026-10-21
- Skip the open access fees: NIH-funded papers must go on PMC with no embargo — anshulkundaje · 2026-10-21
- Creed: LLM code auditing infers developer beliefs to find semantic security flaws — mboehme_ · 2026-10-21
- Spotify removed 50M+ AI slop tracks from recommendations in a year to back human artists — moustaki · 2026-10-21
- METR signs off on Anthropic's risk report but experts demand quantified uncertainty and neutral evals — DKokotajlo · 2026-10-21
- Mistral CEO Arthur Mensch on European AI sovereignty, gigawatt-scale infra and AGI risk — beffjezos · 2026-10-21