Anthropic’s amended S-1 says future models may “imminently” meet its test for AI-accelerated research
The S-1/A, accepted Monday, lists third-quarter revenue of $16.956 billion and a net loss of $23.291 billion. The risk factors say estimates for models that are not yet current indicate the company may soon meet a Responsible Scaling Policy threshold, and that no current model has been designated past it. Purchase commitments are about $895.4 billion.
Feature6 minIn-depth
Priya Shah6 min read
- Anthropic
- IPO
- Safety
- Compute

Photo: AFP / Getty Images, via CNBCSource page
Anthropic told prospective shareholders on Monday that its estimates for models it has not designated as current indicate the company may soon meet an internal test for research sped up by AI. The language is in Amendment No. 1 to its Form S-1. The SEC index for that filing lists an accepted time of 09:24:24 on 2 November. Julie Bort at TechCrunch, in a story timestamped 2:14 p.m. Pacific, 23:14 in Amsterdam, quoted the new risk-factor sentences and described them as a warning that the lab is close to recursive self-improvement. The sentences are more particular than that headline.
For example, 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.” We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data.
— Anthropic, Form S-1/A risk factors, filed 2 November 2026
That sentence is the company’s own test, and it is about models Anthropic does not yet treat as current. The same risk-factor passage says much of the evidence on the pace comes from lagging indicators, and that the ability to forecast the rate, “including any acceleration attributable to our most recent or unreleased models, is limited.” It says Anthropic has in the past failed to identify a significant acceleration until after it occurred, and may fail to do so again. It says the company is evaluating what it would mean to meet “the AI research and development capability threshold” and will release further reports under the Responsible Scaling Policy. Precautionary measures already taken, it says, have included delaying a release and limiting initial availability, “such as with Mythos through Project Glasswing,” and future measures “could include additional measures that affect whether, how, and when we develop or deploy new models.” It says those steps could slow capability gains relative to competitors that do not adopt similar constraints.
A different section, on what the prospectus calls “closing the loop,” uses the phrase recursive self-improvement and says the company does not expect to hand the work to Claude alone. “Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress. This is one of the areas covered by our Responsible Scaling Policy, which sets out our plans to monitor and mitigate the risks as we approach what has been termed ‘recursive self-improvement.’” After a sentence on safety techniques the company says will let it manage those risks, the prospectus continues: “However, we do 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. Accordingly, we are considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.” Bort writes that an S-1 is both a disclosure and a piece of marketing, and that a claim of being close to a model that can build itself can serve the second of those. The document does not name the future model, and it does not give a date.
A third quarter the first prospectus did not include
The amendment is also the first public prospectus table with the quarter ended 30 September. Ashley Capoot at CNBC, in a story published at 9:32 a.m. Eastern and updated at 12:52 p.m. Eastern, noted that the initial filing had stopped at the first half. The quarterly statement of operations lists revenue of $16,956 million for the three months ended 30 September 2026, against $11,555 million in the June quarter and $1,382 million in the year-earlier quarter. The narrative rounds the September figure to $17.0 billion and states year-over-year growth of 1,127 percent for that quarter. Utkarsh Shetti and Arasu Kannagi Basil, in a Reuters story carried by Investing.com, describe the sequential increase as 47 percent, after 144 percent growth in the second quarter. From the June quarter to September, the filing says, higher usage was partly offset by about $600 million of price reductions on new model versions and contractual discounts.
The operating result and the net result are different lines. Operating income was $974 million in the June quarter. The September quarter shows an operating loss of $567 million. The filing says that loss was driven primarily by $5.0 billion of compute bought under short-term commitments, “which currently costs significantly more than comparable compute procured under our multi-year commitments,” and by $2.0 billion of non-cash charges for the donation-matching program. Compute and infrastructure expense was 58 percent of revenue in the September quarter, up from 48 percent in June, and $9,863 million in dollars, which the narrative rounds to $9.9 billion. The net loss for the quarter is $23,291 million, against $66,309 million in the June quarter. The largest single item below operating income is the change in fair value of convertible notes, derivative liabilities, and SAFEs: $21,936 million of expense in the September quarter, and $59,597 million in the June quarter.
For the nine months ended 30 September, the statement lists revenue of $33,242 million, up from $2,545 million a year earlier. The filing says that revenue increased by $30.7 billion, or 1,206 percent. The nine-month net loss is $104,039 million, against $25,107 million in the same stretch of 2025. Of the 2026 nine-month loss, $94,324 million is the fair-value change on those notes, derivatives, and SAFEs. The filing says those marks were driven primarily by a higher enterprise valuation. Compute and infrastructure expense for the nine months rose by $14.4 billion, or 329 percent, to $18,808 million.
Commitments that run past 2040
The prospectus puts purchase commitments and lease agreements, including short-term commitments and leases not yet commenced, at an aggregate of about $895.4 billion as of and after 30 September, “substantially all of which is related to compute and infrastructure.” About $447.5 billion of that is expected to be incurred before 31 December 2030, and substantially all of the rest between 2031 and 2045. About $69.8 billion is due in the 12 months after 30 September 2026. A table of contractual obligations dated 30 September totals $319,993 million on its own. The filing says about $213.8 billion of its contractual obligations relates to arrangements with termination rights, or to arrangements that start only if a counterparty secures financing, facilities, or regulatory approvals. If those conditions are not met, it says, the arrangement may be terminated. Of the $69.8 billion due in the next 12 months, about $4.3 billion sits in that contingent or terminable category.
In addition, certain of our compute providers and infrastructure partners depend on third-party financing or credit support to acquire chips and develop capacity, and if such financing or support becomes unavailable or more costly, contracted capacity may be delayed, repriced, reduced, or not delivered.
— Anthropic, Form S-1/A
Customers above a modest annual bar multiplied during the year. The filing says more than 11,900 customers had $100,000 or more of revenue on a trailing-twelve-month basis as of 30 September 2026, up from about 1,500 at 31 December 2025 and from more than 950 at 30 September 2025. More than 1,900 had $1 million or more, and about 250 had $10 million or more. The remaining 26 percent of trailing-twelve-month revenue, it says, came from about 40 million customers each below $100,000. Shetti and Basil write that Anthropic is preparing to begin formally marketing shares to prospective investors next week. Matt Kennedy, a senior strategist at Renaissance Capital, told them: “I wouldn’t say these results are worrisome, but I do expect some of this pressure on growth and margins to be reflected in the ultimate IPO valuation.” The amendment does not set a share price or a deal size.


