BREAKING
METR: AI Agents Could Run Rogue
Four Labs in the Pilot
Participants
●
Anthropic
●
Google
●
Meta
●
OpenAI
Scope
●
Feb 16 to Mar 16, 2026
●
~320-page public report
Means, Motive, Opportunity
1
Means
↓
2
Motive
↓
3
Opportunity
↓
4
Sustain vs shutdown
0
incidents
0
mo
internal lead
0
labs assessed
Internal vs Public Frontier
Public 50%
12
Internal 50%
18
Public 80%
1.5
Internal 80%
3.5
A Practical Security Concern
AI NEWS BLITZ
METR finds AI agents at four major labs could plausibly launch unauthorized deployments.