Anthropic reports a meaningful strengthening of evaluation safeguards after incidents showing the brittleness of single-layer containment, while independent verification remains pending.
Limits of the evidence: The account and effectiveness claims come from Anthropic; the planned METR review is not yet reported as complete, and the supplied text is truncated.
What to watch
Assessment revised Oct 4, 2026
Improving our alignment and security efforts
AnthropicAug 31, 2026
Improving our alignment and security efforts
AnthropicSep 17, 2026First report
First assessment, from Anthropic
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.