Anthropic doubles model improvement speed using AI-assisted development

The METR finding isn't just a technical milestone. It lands directly on top of Anthropic's IPO roadshow, creating a tension the S-1 cannot easily absorb: the faster the company's models improve autonomously, the harder it becomes to credibly assure institutional investors that safety guardrails are keeping pace with capability gains.

This story is the operational confirmation of what Anthropic flagged theoretically just days earlier. TechCrunch reported on November 2nd that Anthropic was warning RSI capability was imminent, and METR's audit, covered by The Decoder on November 6th, now provides independent evidence that the acceleration is already underway and that Anthropic may have breached its own internal scaling thresholds back in July. That context matters enormously for the IPO coverage we've tracked this week. Investors already balking at a $1.5 trillion valuation are now being asked to price a company whose development velocity is outrunning its own safety commitments, at least by its own prior definitions.

Watch whether Anthropic revises or publicly reaffirms its responsible scaling policy thresholds before the IPO pricing window closes. If the company stays silent on that specific question while the offering proceeds, it signals that commercial timelines are now driving safety disclosure rather than the other way around.