Anthropic Warns AI May Soon Speed Up Its Own Progress

Anthropic’s amended IPO filing carries a warning that its original prospectus did not. In Amendment No. 1 to its Form S-1, filed with the US Securities and Exchange Commission on 2 November, the company says that “our internal estimates and our expectations regarding future models, rather than our current models, indicate that we may imminently meet our Responsible Scaling Policy’s definition of ‘dramatically accelerated progress attributable to AI.’” It adds straight away: “We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data.” TechCrunch was first to point to the passage, which sits deep in the risk factors, well behind the revenue figures that drew most of the attention.

A text search of both documents shows the passage is new. The S-1 Anthropic filed in late October already warned that using Claude to build later Claude models could carry flaws from one generation into the next, but it contained neither the word “imminently” nor the threshold language. The amendment also explains why the company may be late to notice: much of its evidence on the pace of AI-driven acceleration “is derived from lagging indicators,” and it says it has “in the past not identified, and may in the future not identify, a significant acceleration in our rate of development until after it has occurred.” Anthropic says it is evaluating what meeting the AI research and development capability threshold would mean and will publish further reports, as its policy requires.

A second addition is in the business section. Both filings say that “using AI to help build more capable AI could also pose a considerable safety risk” and name “recursive self-improvement” as the point the policy is meant to watch for. The amendment goes on: Anthropic does “not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding, if and when that becomes technically possible,” and so is “considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.” The surrounding numbers explain the concern. The value of Anthropic’s own internal Claude usage at list prices grew more than 250 times between January and September 2026, and in April it first exceeded the company’s monthly personnel expense, excluding stock-based compensation. In June the company said Claude was writing most of its merged code while insisting this was not yet recursive self-improvement.

The threshold itself is defined in Anthropic’s Responsible Scaling Policy. The policy page gives earlier wording as an example, the ability to “compress two years of 2018 – 2024 AI progress into a single year,” and clarifies that this means doubling the rate of progress in aggregate AI capabilities, not doubling the productivity of individual researchers. Version 3.4, in effect since 8 July, revised the automated R&D threshold again, and the page does not reproduce the new text. The filing lists what crossing such lines has meant in practice: delaying releases and limiting initial access, as with Mythos through Project Glasswing. It also warns investors that such measures could become more frequent and costly and could slow Anthropic’s capability gains relative to competitors that do not adopt similar constraints.

The amendment also fills in figures that were blank in October. Revenue for the three months to 30 September was $17.0 billion, up 1,127% year over year, while compute and infrastructure expenses grew 382% over the same period. Headcount was about 5,300, and trailing twelve-month revenue per employee passed $6.6 million, against $1.4 million a year earlier. The pro forma table shows a net loss of $104.0 billion for the first nine months of 2026, which the filing ties in part to changes in the fair value of convertible notes as Anthropic’s valuation rose. The price range is still blank, and the listing remains planned for the Nasdaq Global Select Market under the ticker ANT.

A prospectus serves two audiences, and TechCrunch noted the sceptics’ point that a line about being close to self-building AI can work as marketing as well as disclosure. Risk factors are written to protect the issuer in court as much as to inform buyers. Still, the wording is specific. The filing ties the threshold to a named policy that carries obligations to report, and it warns that the company may only see the crossing after it has happened. Investors pricing the offering will also be pricing in the chance that Anthropic decides to slow down.