METR audit finds Anthropic doubled development pace through AI-assisted research

The timing is the buried detail: if Anthropic crossed its own acceleration threshold in July, that predates both its IPO filing and its public RSI proximity warning, meaning investors and the public received safety assurances after the internal limits may already have been breached.

This lands directly on top of two recent stories. Anthropic's warning that it was close to recursive self-improvement capability (TechCrunch, November 2) now reads differently if internal velocity thresholds were already exceeded months prior, since that warning implicitly assumed guardrails were still operative. Meanwhile, the IPO valuation resistance story from November 6 gets a new dimension: institutional investors already skeptical of a $1.5 trillion price tag now have an independent audit suggesting the company's own safety governance may have slipped without public disclosure. Redacting key evidence from the METR report compounds the problem, because it prevents the kind of third-party verification that would let investors or regulators assess whether the breach was material.

Watch whether the SEC requests unredacted METR audit materials as part of IPO review in the next 30 days. If it does, that confirms regulators view the threshold breach as a disclosure issue, not just an internal safety matter.