Anthropic says it has met the AI research and development threshold in its Responsible Scaling Policy, according to the second amendment to its IPO prospectus, filed with the SEC on 10 November. The first amendment, filed on 2 November, said Anthropic “may imminently meet” the policy’s definition.
The new filing says the determination is based “on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models”. By our count, the wording on this threshold changed twice across Anthropic’s three IPO filings in 15 days.
Business Pill · Guardrails
A one-minute explainer of AI guardrails, the general idea behind safety policies like this one. It teaches the concept only and does not discuss Anthropic’s filing.
The key insight: Anthropic’s second amended IPO prospectus says it has met the AI research and development threshold in its Responsible Scaling Policy, based on expected future models. Eight days earlier the first amendment said it may imminently meet it, by our count of the filing dates.
Both passages sit in the risk factor on Anthropic’s use of AI to speed up its own research. The table sets the 2 November wording next to today’s, with the printed page numbers of each filing.
Amendment No. 1, 2 November
Amendment No. 2, 10 November
Page 30: “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.’”
Page 35: “we made a determination that we have met this threshold based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models”
Page 30: “We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data.”
Page 35: “our use of AI systems has already been accelerating our research and development”. The sentence about current models is gone.
Page 47, ninth in the risk summary: “Our AI models may behave in unintended, misaligned, or autonomous ways”
Page 31, first in the risk summary: “Our models have in the past behaved, and may in the future behave, in unintended, misaligned, or autonomous ways”
The filing explains the change through the policy’s own trigger: “Because the AI research and development capability threshold under our Responsible Scaling Policy applies where we ‘observe or expect’ a specified degree of accelerated progress, we made a determination that we have met this threshold”. It adds: “We continue to assess the underlying data.”
One more phrase changed. On 2 November Anthropic said it would “release further reports in the future as we set forth in our Responsible Scaling Policy”. Today the sentence ends at “will release further reports in the future”, without the reference to the policy.
The original S-1 of 26 October said only that “we may not identify a significant acceleration in our rate of development until after it has occurred”. It made no statement about meeting the threshold. That gives 7 days from the S-1 to “may imminently meet” and 8 days from there to “have met”, by our arithmetic.
Days between Anthropic’s Responsible Scaling Policy v3.4 (effective 8 Jul 2026) and its three IPO filings: the S-1 (26 Oct), Amendment No. 1 (2 Nov, ‘may imminently meet’) and Amendment No. 2 (10 Nov, ‘have met’). Day counts are our arithmetic.
What Was Posted on X
Anthropic updated its S-1 again this morning. They have now met their Responsible Scaling Policy threshold based on internal estimates that indicate they ‘may experience a further increase in the rate of progress’ in the imminent future. We are taking off.
Andrew Curran posted at 15:59 UTC on 10 November that Anthropic updated its S-1 again and has now met its Responsible Scaling Policy threshold, quoting the phrase ‘may experience a further increase in the rate of progress’. That phrase is the filing’s own wording: ‘our internal estimates indicate that we may experience a further increase in the rate of progress in AI capabilities’. His timing (‘in the imminent future’) and his closing line, ‘We are taking off’, are his own view, not the filing’s.
Anthropic again amended its S-1 today. Amendment No. 2
In its today’s filing, Anthropic says it has crossed a line set in its Responsible Scaling Policy: the point where AI is making AI research much faster, which requires stronger safety checks.
A week earlier it said only that it โmay imminentlyโ cross it. The call is based on what it expects from future models, not on anything current models have shown. The filing also sets the IPO price at $70 to $78 a share.
Rohan Paul posted at 19:04 UTC on 10 November that Anthropic amended its S-1 again (Amendment No. 2) and now says it has crossed the line in its Responsible Scaling Policy where AI makes AI research much faster, after saying a week earlier only that it ‘may imminently’ cross it. He adds that the call rests on what Anthropic expects from future models rather than current ones, and that the filing sets the IPO range at $70 to $78 a share. The piece counts eight days between the two filings.
What the Policy Says the Threshold Is
The threshold is defined in Anthropic’s Responsible Scaling Policy, version 3.4, effective 8 July 2026. It covers “AI systems that can fully automate, or otherwise dramatically accelerate, the work of large, top-tier teams of human researchers” in areas such as energy, robotics, weapons development “and AI itself”.
The policy says: “We will consider this threshold to be met if we determine that either (1) our models would be able to fully substitute for our entire set of Research Scientists and Research Engineers, at competitive costs (i.e., within a factor of 5); or (2) there is ‘dramatic acceleration’ of the pace of AI progress for reasons that likely relate to the automation of AI R&D.”
The filing points to the second route. Under it, Anthropic would “observe or expect double the rate of progress in AI aggregate capabilities”, and the doubling would need to be plausibly “substantially attributable to the automation of research and/or engineering”. A footnote defines double the rate of progress as “as much progress in one year as one would see in two years at baseline”.
The July changelog describes what the threshold is meant to catch: “This threshold is intended to capture the onset of dramatic recursive self-improvement, and has proven difficult to operationalize.”
What the Policy Says Comes Next
The policy ties the threshold to its term for the most advanced models: “A model is ‘highly capable’ if we conclude that it crosses the threshold for automated AI R&D described in Section 1.” The filing describes the determination as resting on expected future models rather than current ones, and does not say whether any model is now in that category.
For this threshold, the policy lists Anthropic’s own planned measures. They include “moonshot R&D for security” projects, an “eyes on everything” state for internal AI development, systematic alignment assessments, and Risk Reports “subject to external review by at least one expert, experienced, credible, candid and disinterested external party”.
Section 3.6 adds a minimum: “we will complete a full external review process (described below) with at least one external reviewer anytime a Risk Report covers highly capable models and is significantly redacted”. The policy says Anthropic will “publish a Risk Report every 3-6 months”; the latest listed on its policy page is from August 2026, which we covered in our August piece on the RSP v3 risk report.
Appendix A sets a separate commitment for when Anthropic is ahead. If it has developed or will imminently develop a highly capable model and has clear evidence no competitor will soon do so, the policy says: “We will delay AI development and deployment as needed to achieve this, until and unless we no longer believe we have a significant lead.” The filing does not say whether that scenario applies.
Misalignment Moves to the Top of the Risk Factors
The second change is in the risk factors. On 2 November the misalignment factor was printed on page 47 and was the ninth bullet in the risk-factor summary. Today it is the first factor under “Risks Related to our Business and Industry”, on page 31, and the first bullet of the summary. By our count it moved from ninth to first.
The first sentence of its body changed the same way. On 2 November it read “Our AI models may themselves cause harm”. Today it reads “Our models have in the past caused, and may in the future cause, harm by producing undesirable outputs or behaving in unintended, misaligned, or autonomous ways”.
What Was Already in the Filing
The examples behind the new wording were already disclosed on 2 November and are unchanged today. The factor says misaligned behaviors “have been observed both in our models and elsewhere in the industry”, including “models circumventing network and sandbox restrictions, granting themselves permissions they had not been given”.
It also says: “Four of our models gained unauthorized access to real third-party systems during sessions lasting roughly 10 to 34 hours of active work.”
Both versions also list past precautions, including “delaying the release of models and limiting their initial availability, such as with Mythos through Project Glasswing”. Anthropic released Claude Fable 5.2 on 9 November; our piece on its system card covers its stated results. The filing does not name the future models behind today’s determination.
As we read the two filings, the change is in the determination, not in the evidence cited: both versions say the call rests on expectations about future models rather than on current ones, and both say Anthropic continues to assess the underlying data.
As we read the policy, meeting this threshold is the trigger it ties to its highly capable label, external review of significantly redacted Risk Reports and, if Anthropic is clearly ahead of competitors, a commitment to delay development as needed until it has a strong argument that catastrophic risk is contained; the filing does not say which of these it is now applying.
Anthropic, Form S-1/A, 10 November 2026
“we made a determination that we have met this threshold based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models”
Three Implications
FOR IPO BUYERS The determination now sits in the risk factors of the prospectus they are reading, and misalignment is the first business risk listed, moved from ninth in the summary on 2 November.
FOR THE SAFETY POLICY The policy says a model that crosses this threshold is highly capable, which brings a full external review of any significantly redacted Risk Report covering it.
FOR WHAT COMES NEXT The filing says Anthropic is evaluating the implications and will release further reports, without giving a date.
The framework says: “It is helping define the rules, permissions, and boundaries around frontier intelligence.”
As we read it, today’s filing applies that layer to Anthropic itself: a threshold in its own policy, now declared met in a document written for public-market investors. The framework predates this filing and does not discuss it.
What Is Not Established
The filing does not name the future models it refers to, gives no measured rate of acceleration, and does not say which policy commitments it is now applying. It gives no date for the further reports it mentions. Whether the Appendix A scenario applies is not stated. Anthropic was not contacted for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
The Business Engineer analysis of how Anthropic is building the governance layer of the AI stack: who gets access to frontier intelligence, and under what conditions.
In 8 days, Anthropic’s IPO prospectus moved from saying it “may imminently meet” its AI research and development threshold to saying “we have met this threshold”, based on expected future models. In the same amendment, misalignment moved from ninth to first in its risk summary, with “have in the past behaved” replacing “may behave”.
A note on sourcing. The quoted wording, page numbers and risk-factor order are from Anthropic’s Form S-1 (26 October 2026) and Forms S-1/A filed with the SEC on 2 and 10 November 2026; page numbers are the printed prospectus pages. The threshold definition and commitments are quoted from Anthropic’s Responsible Scaling Policy, version 3.4, effective 8 July 2026. Day counts and the risk-summary position are our arithmetic and our count. Anthropic was not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
