Anthropic Says Its AI Crossed a Key Safety Threshold Before Its IPO
Anthropic Says Its AI Crossed a Key Safety Threshold Before Its IPO
What the Threshold Means
A week before its initial public offering, Anthropic warned investors that AI speeding up its own research might "imminently" trigger one of its key safety alarms. On Tuesday, the company said it already had.
The announcement came in an amended filing that also set a price range for the offering. The company wrote that it had "made a determination that we have met this threshold." The call, it said, rested on "what we expect from future models and our overall rate of progress," not on what it had seen from current models.
"We made a determination that we have met this threshold."
Here is what that means in plain terms. Anthropic's Responsible Scaling Policy is the company's internal rulebook for adding safety measures as its models grow more capable. One of its tiers applies when AI does enough of the company's research to roughly double the pace at which its models improve. The policy's own math says that pace would pack two years of AI progress, measured at the 2018 to 2024 rate, into a single year.
Crossing the line does not shut anything down. It means more safeguards. The filing says AI "has already been accelerating our research and development," and the company says it is weighing policy changes to keep humans in control.
Why the Speedup Worries People
The concern in the field is simple. If AI improves itself faster than people can check its work, human oversight starts to lag behind the thing it is supposed to supervise. In 2023, Anthropic CEO Dario Amodei joined other AI leaders in warning that the risk of extinction from AI should be treated as a global priority, on par with pandemics and nuclear war.
Anthropic says humans stay in charge. The filing says the company does not expect to use its models to design and build successive generations of frontier systems without meaningful human oversight.
"Without meaningful human supervision or understanding."
The Outside Critique
The sharper claim comes from outside the company. 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." METR said it was plausible the line was crossed by mid-July.
That is a timing dispute, and it matters. If METR is right, the threshold was crossed months before the company said so, and the August argument did not hold up. If Anthropic is right, the speedup is still building and the company is simply moving early. Both readings raise the same question for investors: how much of the company's own forecasting can they trust?
A $1.4 Trillion Debut
The money side moves quickly. Anthropic set its price range at $70 to $78 per share on Tuesday, the start of a roadshow of investor meetings before pricing. At the top of that range, the company would be valued at about $1.4 trillion, counting every share that could exist. Reports say trading could begin as soon as November 19.
What Readers Should Watch
Nothing here is secret. The filing is public, and METR published its assessment. The unusual part is the timing, with the disclosure arriving in the same document that set the price range.
Investors are being asked to buy into a company whose own filing says its technology is accelerating its own development, and whose outside evaluator says the company may have been late to say so. Read the filing directly, and treat both the company's claims and METR's critique as claims to check.
Aaron Rose is the founder and publisher of Tech Reader Magazine, covering system architecture, cloud platforms, and AI policy.