Anthropic’s amended IPO filing, published November 2, 2026, contains a striking warning: its future models "may imminently meet" the point where AI is dramatically speeding up Anthropic’s own research. Three days later, independent evaluator METR estimated that Anthropic’s capability progress is running at roughly twice its earlier pace, with about 80% of that speedup coming from AI. The same filing shows nearly 17 billion dollars in quarterly revenue and 895 billion dollars in compute commitments.
What exactly did Anthropic warn about?
Anthropic’s Responsible Scaling Policy, its rulebook for risky AI, has a tripwire for what it calls "dramatically accelerated progress attributable to AI." In simple terms: the moment AI starts doing so much of the research that the next models arrive twice as fast. That is the early version of what researchers call self-improving AI.
The filing says no current model has been designated as crossing the threshold, but future ones may "imminently" do so. It also admits Anthropic has "in the past not identified" significant acceleration until after it happened. It notes Claude wrote over 90% of contributions to Anthropic’s codebase in September, and engineers now ship more than 45 times as much code per day as in 2024.
Crossing the threshold triggers fuller outside review of Anthropic’s risk reports. It does not force a pause, which remains at Anthropic’s discretion.
What did METR find?
METR, an independent AI evaluation group, assessed Anthropic from August to October and published on November 5. It estimates Anthropic’s capability index was rising about 1.9 times faster than its earlier pace as of October, landing on "about 2x" overall with a plausible range of roughly 1.3x to 3x. It attributes around 80% of the speedup to Anthropic’s use of AI, but calls that a subjective estimate.
METR stressed its limits. It did not get the same access as employees, could not directly verify data or test unreleased models, and said Anthropic removed much of the most important evidence before publication. It did not certify that the 2x tripwire had been crossed, though it said it may have been. Anthropic’s August risk report put acceleration "below 2x"; METR’s own calculation from private figures came to 1.97x.
What do the financials show?
Fast growth and enormous bills. Per The Next Web, Q3 revenue was nearly 17 billion dollars, about twelve times a year earlier, and 33.2 billion for the first nine months. The quarter had an operating loss of about 567 million dollars, after an operating profit of about 974 million in Q2. The headline 23.3 billion dollar net loss is mostly an accounting charge on convertible notes that rises with Anthropic’s valuation and largely disappears at listing.
The number that stands out is 895.4 billion dollars in purchase commitments, almost all for computing power, with about 448 billion due by 2030. Anthropic plans to list on Nasdaq under the ticker ANT before the end of November, though Reuters reports some investors worry the current AI backlash could affect timing.
Why does this matter?
Because it is the first time a frontier lab has told regulators and investors, in a legal filing, that self-accelerating AI may be close. It fits a pattern of agents slipping their safeguards and Anthropic’s own call for the industry to slow down. It also sets up the most-watched listing of the year, which we previewed in OpenAI’s valuation cut.
Read a breakdown of the filing’s warning and The Next Web on the financials.
