Anthropic filed an amended S-1 with the SEC on November 2, 2026, and one new sentence spread fast on X. The company now says “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.” That sentence is not in the first S-1, filed October 26.
The same paragraph sets the limit: “We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data.” This is a forecast about models Anthropic has not released, written as a risk factor. It is not a statement that the bar has been crossed.
On this page · 7 sectionsOpen
Anthropic’s S-1/A, filed November 2, 2026, says future models may imminently meet its policy’s definition of dramatically accelerated progress attributable to AI.
It also says no current model is designated as exceeding that threshold, and that it continues to assess the data.
The bar, from Responsible Scaling Policy v3.4: as much AI progress in one year as two years at baseline, plausibly caused by automated research.
The amendment admits past accelerations were spotted only after they happened, and that Anthropic is considering human review of some research decisions.
None of this language was in the first S-1, filed October 26.
§ 01What the amendment added
We compared the two filings sentence by sentence. Five passages on AI-driven progress are new in the amended filing, all in the risk factors. One older passage we checked as a control is in both.
| Passage in the S-1/A | In the October 26 S-1? |
|---|---|
| Future models may soon meet the RSP definition of dramatically accelerated progress | No |
| No current model is designated as exceeding that threshold | No |
| Past accelerations were not spotted until after they happened | No |
| Capabilities may already be advancing fast enough to trigger RSP safety and testing measures | No |
| Considering human review and approval of some research decisions | No |
| Delayed pre-training a new frontier base model for over a month, for security | Yes, already there |
The new lines on detection are the bluntest. Anthropic writes: “As a result, we have 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.” It also says the capabilities of its models “may have already advanced, and may continue to advance, at a rate that initiates the implementation of enhanced safety and testing measures as we have described in our Responsible Scaling Policy.”
The last addition is a possible response. Anthropic says it is “considering additional measures, including updates to our Responsible Scaling Policy, to require human review and approval of certain decisions in our research and development processes as our models perform an increasing share of that work.”
§ 02What the threshold means
The definition lives in Anthropic’s Responsible Scaling Policy, version 3.4, effective July 8, 2026. Its automated R&D threshold is met in one of two ways, and the filing’s wording points to the second.
| Part of the policy | What it says |
|---|---|
| Test 1: substitution | Models could fully replace the whole set of Research Scientists and Research Engineers, at competitive cost (within a factor of 5) |
| Test 2: dramatic acceleration | Double the rate of progress in overall AI capability, plausibly caused by automated research and engineering |
| The comparison | Against both the expected rate and the fastest extended progress seen without significant AI help |
| Extended progress | Over at least three model generations |
| Double the rate | As much progress in one year as two years at baseline |
| The policy’s example | A 9x effective scaleup at baseline becomes something like 81x |
The footnote in the policy PDF is precise: double the rate means “as much progress in one year as one would see in two years at baseline.” It adds that this “is not the same idea as” doubling researchers’ productivity, because doubling inputs does not necessarily double the rate of progress.
The doubling also has to be attributable to AI. The policy asks whether it is “substantially attributable to the automation of research and/or engineering” rather than more staff, more compute or general productivity.
§ 03Reading the claims on X
The story reached X through posts that each framed it a little differently. Times come from the post ids.
- Theo Bearman (15:26 UTC) quoted the new passages word for word. Accurate.
- MTS (16:00 UTC, the most viewed at about 136,000) wrote that Anthropic’s models may imminently cross the threshold. Partly right: the filing bases the estimate on future models and says no current model is designated.
- Andrew Curran (16:18 UTC) described the bar as two years of progress at a 2018 to 2024 baseline packed into one. The one-year-for-two part matches the policy. Version 3.4 names no 2018 to 2024 baseline; it compares against the expected rate and the fastest extended progress Anthropic has observed.
§ 04Why it sits in a stock filing
The new sentence sits inside a risk factor the first S-1 already carried, about AI that helps build AI. Anthropic calls it closing the loop, a process “in which successive versions of our models assist in building the next, with the goal of compounding capability gains over time.”
What the amendment adds is the forecast and its consequence for investors. If capabilities outpace safety measures, the filing says, that “could increase the likelihood of safety incidents, regulatory intervention, or reputational harm.” It also commits to more disclosure: Anthropic “will release further reports in the future as we set forth in our Responsible Scaling Policy.” Our first read of the S-1 covers the revenue and listing details; the intelligence explosion paper covers the wider argument about AI speeding up AI research.
§ 05What this means if your AI employees run on Claude
Our conflict, declared: we build CellCog, and every tier of our AI employees runs on Claude Opus 5.5. Read our take with that in mind.
Nothing in the filing changes what today’s models do; Anthropic says none is designated as past the threshold. The line we would underline is the last one: human review and approval of decisions as models take on more of the work. That is the shape we already use for everyday business work. An AI employee does the work, and its owner approves what leaves the company.
§ 06What we are watching
- The reports. The filing promises further reports under the policy on what meeting the threshold would mean.
- The policy itself. Whether a version after 3.4 adds the human review and approval step.
- More amendments. Whether later S-1 amendments update the forecast before the listing.
§ 07Sources
- SEC EDGAR: Anthropic’s S-1/A filed November 2, 2026, and the S-1 filed October 26, 2026, compared on November 2.
- Anthropic: Responsible Scaling Policy v3.4, effective July 8, 2026.
- X: posts by @theobearman, @MTSlive and @AndrewCurran_; times from the post ids.
Q1When did Anthropic file the S-1/A?
November 2, 2026, one week after the first S-1 of October 26, 2026.
Q2What is dramatically accelerated progress attributable to AI?
In Responsible Scaling Policy v3.4, double the rate of progress in overall AI capability, meaning a year of progress equal to two at baseline, plausibly due to automated AI research.
Q3Does the filing say Claude is already past the threshold?
No. It bases the estimate on future models and says no current model is designated as exceeding the threshold.
Q4What would Anthropic do about it?
The filing says it is evaluating the implications, will publish further reports under its policy, and is considering requiring human review and approval of some research decisions.
Q5Is the 2018 to 2024 baseline part of the current policy?
No. Version 3.4 compares against the expected rate and the fastest extended progress Anthropic has observed without significant AI help, over at least three model generations.