METR says Anthropic may have crossed its own AI R&D warning threshold by July

Reviewing Anthropic's August risk report with non-public access, METR agrees that catastrophic risk from models as of July 15 was low, but says it is plausible they had crossed the company's early warning threshold and that the public report understates the acceleration. METR's staff and Anthropic's both lean toward it not having been crossed. Anthropic had redaction rights.

Why it matters: It is a rare outside check on whether AI is speeding up AI research, directly at a frontier lab.