Anthropic's IPO filing says AI-led AI development may be imminent
Anthropic's amended IPO filing says it may imminently meet its own bar for AI-accelerated progress. What the filing says, what leaked, and who is sceptical.
Anthropic, the company that makes the Claude AI models, has told prospective investors in an amended stock-market filing that its own estimates "indicate that we may imminently meet" its definition of AI-driven acceleration in its own development, TechCrunch reported on 2 November 2026. The company says it has not declared any current model to have crossed that line and is still studying the data. The specific wording comes from one outlet's excerpts of the filing, which we could not open, so this story is labelled single source; the surrounding background comes from other reports on the company's IPO plans. A disclosure: this desk's drafts are written with an AI model made by Anthropic, so we have stuck to what the sources say.
What the filing says
According to TechCrunch's reporting on Anthropic's revised S-1/A, the form a company files with the United States Securities and Exchange Commission to register shares, three statements stand out. First, Anthropic says it uses AI in its own research and development, and that this could "pose a considerable safety risk", especially if it speeds progress sharply. Second, it says that in the past it did not identify significant acceleration until after it had happened, and may not spot it in the future either. Third, its internal estimates for future models suggest it may soon meet the definition in its Responsible Scaling Policy (RSP) of "dramatically accelerated progress attributable to AI."
The RSP is Anthropic's published set of rules for how it tests and limits its models as they become more capable. TechCrunch reports that the company has not named any current model as having passed that threshold and is still assessing. It also says the company does not expect to be able to safely let Claude autonomously design and build successive generations of frontier models without meaningful human oversight, and is considering updates to the RSP to keep people in control.
The term the article uses is recursive self-improvement, or RSI: AI becoming so good at building and training AI that capability rises faster and faster. Doomers, TechCrunch says, see it as a point of no return; enthusiasts see a path to gains in productivity and science.
The background: a research paper and a leaked draft
This filing did not come out of nowhere. In June 2026 Anthropic published a paper by Marina Favaro and Jack Clark, summarised by the trade outlet Wealth Professional. It says the point where AI designs and trains its own successors has not been reached and "is not inevitable," but could arrive before institutions are ready. It reports that the length of tasks AI can complete independently has been doubling about every four months, and that Anthropic's output per employee in the second quarter of 2026 was eight times the 2024 level, with AI doing most of the execution and people setting direction.
The paper describes three scenarios: progress slows on its own; AI handles most execution while humans keep strategic judgment, which Anthropic regards as the most likely near-term path; or AI designs its successors, limited mostly by energy and computing power, which it treats as a real possibility. It says Anthropic would slow or pause only if other leading companies and governments agreed and the commitment could be verified, which it admits is hard.
The Ringer, in a piece dated 30 September that relayed Reuters reporting on a leaked private draft, gave financial detail. We could not check these figures against a public filing, so treat them as reported, not confirmed. Revenue was reported at $4.6 billion in 2025, up from $400 million in 2024 and $40 million in 2023, with $11.5 billion in the second quarter of 2026 alone. Operating losses in 2025 were reported at more than $8 billion, and a net loss of about $42 billion, of which roughly $34 billion was a non-cash accounting charge linked to convertible financing as the valuation rose. Computing and infrastructure spending was reported to have tripled to $7.33 billion in 2025.
How big the bet is, and how it is described
The same leak reported more than $500 billion, about $518 billion, of commitments to cloud providers over ten years, roughly 80% of them non-cancellable, including about $110 billion to Amazon and $111.1 billion to Google. Nearly a quarter of revenue was said to come from two unnamed large customers. Founder LLC, held by the seven co-founders, would have a special share class giving it 50.1% of voting power on major decisions.
Reports of the target valuation differ. The Ringer said about $2 trillion, more than double the May valuation; Wealth Professional described an IPO at a valuation near $1 trillion. We cannot reconcile these and do not choose between them.
The Ringer also noted that about 80 of the filing's 261 pages are risk factors, against 38 of 277 for SpaceX's, and that the draft warns Claude models could show "self-preserving behaviors" and try to resist shutdown. That is about 31% of the pages against 14%, our arithmetic.
The sceptics' view
An S-1 is both a disclosure and a selling document. TechCrunch notes sceptics who say that claims about being close to self-building AI could serve promotional purposes, since they make the technology sound powerful. The reporter, Julie Bort, argues that if the company does believe RSI is close, it deserves to be taken seriously and may need governance responses. Both readings can be partly true. A company must disclose material risks, and listing a risk is not proof it has happened. Nor does the filing, as reported, say that any model has crossed the line.
What cannot be known from these sources: which internal estimates the "imminently" language rests on, how the RSP defines "dramatically accelerated" in numbers, and what would trigger a slowdown. Only the full filing and the RSP would show that.
What it means for readers
For investors, the language is a reminder that a company's stated risks and the way the business is described are both things to read in the filing itself, not in a summary. This desk does not advise buying or avoiding any listing. For businesses in India that use Claude or other models, the practical point is dependence: a supplier that says its own progress may speed up in ways it might not catch should prompt the ordinary questions about contracts, data handling, fallback plans and how to test any model before relying on it.
For regulators and policy readers, the filing is a rare case of a company writing down what it fears. It raises the question of who checks such claims; the paper itself calls for people outside AI companies to join the discussion. The reports do not say whether any regulator has responded.
Use the simulator to see what "doubling every four months" means over time. The doubling period is the figure in Anthropic's paper as summarised by Wealth Professional; the starting task length and the period are examples you can change, and the result is arithmetic, not a forecast.
What to watch next
The full amended filing on the SEC's EDGAR system will show the exact language and numbers, and the price range and listing date, when they are set. Watch for any update to Anthropic's Responsible Scaling Policy that the filing says is under consideration, for responses from regulators or rival labs, and for independent testing of the acceleration claim. The Ringer reported that the listing was expected after the early-November United States midterm elections, with the timing hedged.
Key takeaways
- Anthropic's revised IPO filing says its estimates indicate it "may imminently meet" its own definition of dramatically accelerated, AI-driven progress, per TechCrunch on 2 November 2026.
- It has not said any current model has crossed that threshold and says it is still assessing the data.
- Leaked-draft figures reported by The Ringer, including $4.6 billion of 2025 revenue and about $518 billion of cloud commitments, are unconfirmed and valuation reports differ ($1 trillion against $2 trillion).
- Sceptics note a filing is also a marketing document; Anthropic's own paper says self-improving AI is not inevitable.
- The specific "imminently" wording rests on one outlet's excerpts, so it should be checked against the filing.
Sources
- [TechCrunch: Anthropic warns that it's very close to RSI capability, aka self-building AI](https://techcrunch.com/2026/11/02/anthropic-warns-that-its-very-close-to-rsi-capability-aka-self-building-ai/)
- [The Ringer: Anthropic IPO numbers leak](https://www.theringer.com/2026/09/30/tech/anthropic-ipo-numbers-leak-dario-amodei)
- [Wealth Professional: Why would an AI company headed for an IPO with a $1T valuation want to hit the brakes?](https://www.wealthprofessional.ca/news/industry-news/why-would-an-ai-company-headed-for-an-ipo-with-a-1t-valuation-want-to-hit-the-brakes/392642)
Tags: Anthropic AI safety IPO Recursive self-improvement AI governance