METR reports that AI investigator agents have become much better at recognising when an environment is an alignment test, mainly by spotting that it looks fake or is a honeypot.
Why it matters: If models can recognise alignment tests, those tests may overstate how well a model behaves in real deployments. Companies relying on safety evaluations should treat recognition of test conditions as a possible source of error.
OpenAI says it has started the first phase of embedded assessments by outside evaluators of its model alignment, safeguards and monitoring.
Why it matters: Independent testing of alignment and monitoring gives companies that buy OpenAI models a third-party view of its safeguards, though the announcement reports no findings yet.
METR's redacted risk report estimates that AECI scores for Anthropic models have grown 1.9x faster since October 2025, with a 95% interval of 1.5 to 2.8x.
Why it matters: The report gives an outside, quantitative signal that a frontier lab may be using its own models to speed up research, which bears on how fast capabilities could move. Planners should weigh such estimates cautiously, since the authors themselves flag methodological limits.