Anthropic’s IPO filing says a key safety threshold is met

Anthropic said on November 10, in the amended prospectus that set the price range for its IPO, that AI speeding up its own research has crossed one of the safety thresholds in its Responsible Scaling Policy. The determination rests on what Anthropic expects from future models and its overall rate of progress, not on anything current models have shown, and crossing the line commits the company to tougher safeguards rather than a shutdown.

“We made a determination that we have met this threshold,” the company wrote, as Yahoo Finance reported. A week earlier, Anthropic had told would-be investors the trigger might come “imminently.” The filing says AI “has already been accelerating our research and development.”

The threshold trips when AI does enough of Anthropic’s research to roughly double how fast its models improve. The policy describes that pace as two years of AI progress at its 2018-2024 rate packed into a single year. The fear behind it is AI improving itself faster than people can check it. Anthropic revised this automated R&D threshold in version 3.4 of the policy, effective July 8, to better track that threat model.

The same filing set the price range at $70 to $78 per share as Anthropic began its roadshow, the round of investor meetings before pricing, valuing it at as much as about $1.4 trillion counting every share that could exist. Trading could start as soon as November 19, according to reports.

What METR saw

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 its audit, which ran from August 12 to October 29, the measured part was AECI, Anthropic’s version of the Epoch Capabilities Index, where scores grew 1.9 times the pre-October 2025 rate (95% confidence interval 1.5-2.8x).

In a separate review, METR called Anthropic’s August argument that it was still below the threshold “overstated” and said it was “plausible” the line was already crossed by mid-July. That argument came in the August Risk Report, shared on August 14, which covered risks up to July 15. The overall twofold estimate is METR’s own subjective call, one it said would not be surprising at 1.3x or 3x, and Anthropic redacted most of the evidence METR found most informative.

What the threshold triggers

In the first two versions of the Responsible Scaling Policy, crossing a threshold made the matching safeguards a precondition, and Anthropic committed not to train or deploy until they were in place. Version 3.0, effective February 24, changed that. A pause is now tied to competitors: Appendix A commits Anthropic to delay development and deployment as needed when it has clear evidence it is in the lead, and when every competitor with a comparably capable model can make a strong case that catastrophic risk is contained. The stated reason is a collective action problem, in which one developer pausing while others ship without strong mitigations could leave the world less safe. A later update states that Anthropic remains free to pause in any circumstances it deems appropriate, even where the policy does not require it.

Anthropic says humans stay in charge. The filing 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,” and that it is weighing policy changes “to keep humans in control.”