AI Forecast Ledger™

Third-party dispatches, not graded receipts

Anthropic gives METR evaluators "employee-like" access to internal systems

theverge.com · 2026-10-16

Six METR researchers will get access to hardware and internal tools roughly on par with Anthropic's own risk assessment team, though Anthropic can still require redactions in published reports. This is a notable expansion of independent safety evaluation access at a frontier lab.

“Six employees at the AI safety organization METR will gain access to hardware and internal tools “roughly on par” with what Anthropic’s own risk assessment team sees.”

source↗ · archived copy↗

← all dispatches