Landmark new METR report: Can AIs already start ‘rogue deployments’ inside AI companies?
Rationale
This EA Forum post discusses METR's landmark report on "rogue deployment" risks from internal AI agents at frontier labs, highlighting red-team findings at Anthropic that revealed exploitable monitoring gaps and permission vulnerabilities. The content demonstrates high thesis fit (ai-safety) with recency=1, thesis_fit=1, talent_density=1 (METR researchers Hjalmar Wijk and Ajeya Cotra, collaboration with OpenAI/DeepMind/Meta/Anthropic), and founder_credential=1. However, this is a research dissemination artifact (80,000 Hours podcast transcript reposted to EA Forum), not an organization seeking philanthropic capital, yielding raise_stage=0.5 and composite score 85/100 despite strong alignment signals.
Project Facts
Contacts
Contacts at —
Contact lookup pending owner identification.
HubSpot
Outreach
Verifier — Unverified
unverifiedorg_exists_failed:no_corroborating_field
Source Record
Activity timeline
- Ranked · score 85Email Ajeya Cotra at METR to confirm whether METR is raising philanthropic capital for their next evaluation round referenced in the report (scheduled "in just a few months") and request their current funding deck if active.2026-05-21 18:09 · projects
- Project ingestedSource: custom-ea-forum-rss · posted 2026-05-202026-05-21 05:43 · projects