Anthropic says AI now speeds up its own research, and discloses it to investors in its IPO filing

Anthropic says it has crossed one of its key safety thresholds. Under its Responsible Scaling Policy, the company's rulebook for adding safeguards as its AI gets more capable, this threshold is triggered when AI does enough of Anthropic's own research to roughly double how fast its models improve. The policy describes that as fitting two years of progress, at the 2018–2024 pace, into one. 'We made a determination that we have met this threshold,' Anthropic wrote in its amended prospectus, Yahoo Finance reports. The same filing set an IPO price range of $70 to $78 per share, valuing the company at up to about $1.4 trillion.

The wording is precise. Anthropic says the decision rests on 'what we expect from future models and our overall rate of progress, rather than on what we have observed from current models.' A week earlier, it had told would-be investors the trigger could come 'imminently.' Crossing the threshold means more safeguards, not a shutdown. The company says it does not expect to safely use Claude 'to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding.' It says it is weighing policy changes 'to keep humans in control.'

The outside view matters as much as the announcement. METR, an independent nonprofit that tests AI systems, estimated last week that Anthropic's models have been improving nearly twice as fast since October 2025. In a separate review, METR called Anthropic's August argument that it was still below the threshold 'overstated.' It said it was 'plausible' the line had been crossed by mid-July. Anthropic has not conceded that point. Still, an outside evaluator questioning a lab's own reading of its safety policy, in public, is the kind of oversight that makes these policies worth more than PR.

Our thesis: this is the first time a company's voluntary safety threshold has turned into a disclosure in a securities filing. Once it is in a prospectus, it carries legal responsibility. Investors, regulators and plaintiffs can hold Anthropic to those words. That is good for oversight. It also creates an incentive problem: when announcing a trigger can move a $1.4 trillion valuation, both the timing and the wording become strategic. Basing the call on expected progress rather than observed capability is defensible as caution. It also makes the call hard for outsiders to check, which is why METR's independent measurements matter.

Our reading: we should keep the fear and the opportunity in proportion. The fear, which Dario Amodei signed onto in 2023 when he placed extinction risk alongside pandemics and nuclear war, is AI improving itself faster than people can check it. The opportunity is the same mechanism. Faster AI-driven research is how we get to faster drug discovery, longer healthy lives and broad abundance. The question is not whether research speeds up. It already has. The question is whether oversight keeps pace: independent evaluators with real access, decisions that can be audited, and safeguards that hold when the stock price is under pressure. Anthropic has said where the threshold is. Now it has to show what the tougher safeguards actually are.