METR is proposing a potentially useful transparency mechanism for evaluating internal loss-of-control risks, but the proposal leaves important access and reporting limitations unresolved.
Limits of the evidence: This is METR's own proposal; the source does not establish developer participation, the access companies will grant, or resulting safety outcomes.
What to watch
Assessment revised Oct 15, 2026
A proposal for better informing the public about loss-of-control risks from frontier AI development
METR ResearchOct 14, 2026First report
Update from METR Research
First assessment, from METR Research
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.