METR: Anthropic may have crossed its own AI acceleration warning threshold in July

What the audit found

Nonprofit research organization METR spent eleven weeks reviewing Anthropic's internal research data. Its conclusion: the pace of model development at the company has roughly doubled. About 80 percent of that speedup is driven by AI-assisted research work.

Threshold and transparency concerns

METR says the acceleration may have crossed Anthropic's own AI acceleration warning threshold back in July. However, the company redacted the most informative evidence. As a result, the public has to rely on METR's judgment "much more than would be ideal."

Skepticism inside the company

According to the report, there are also skeptical voices inside Anthropic about the assessment. The acceleration finding is thus a subject of both independent audit and internal debate.

Why it matters

Whether AI development speed is entering a self-reinforcing loop sits at the center of safety and oversight debates. METR's finding can be read as a concrete signal that a leading AI company is approaching or has crossed its own warning threshold.