S-1/A, filed November 2, 2026

Anthropic says it may imminently hit its own threshold for AI speeding up AI

The biggest numbers in the amended filing are financial. The sentence people will return to is in the risk factors, and it wasn't there a week ago.

"...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, amended Form S-1, risk factors

The short version

Anthropic has a safety policy with a line in it: AI making AI progress run about twice as fast as it otherwise would. Its IPO filing now says estimates about future models suggest it may reach that line soon. No current model has been designated past it, and in August the company measured progress below it. The filing also warns that a speedup might only be visible after it has happened.

What happened

Anthropic made its S-1 public on October 26, 2026 and filed an amended version, the S-1/A, on the morning of November 2. Most coverage went to the third quarter numbers: about $17 billion in revenue and about $895 billion in compute and infrastructure commitments, as CNBC reported.

The amendment also added language about AI building AI. A text search for "may imminently meet" finds zero matches in the October 26 filing and one in the November 2 filing. Three other passages are new too:

  • "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."
  • "we are considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic."
  • "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's worth reading this as what it is: a risk factor in a securities filing, where companies are expected to disclose what could go wrong. It's also a dated, legally reviewed statement from a frontier lab, which is why it carries weight.

What the threshold means

The Responsible Scaling Policy (RSP) is Anthropic's rulebook for what safeguards it owes at each level of AI capability. Version 3.4 says its AI research threshold is met in either of two ways. Models could fully substitute for all of its research scientists and engineers at competitive cost. Or there could be "dramatic acceleration" of AI progress, which has two parts:

  1. Anthropic observes or expects double the rate of progress in aggregate AI capabilities, compared with both what it would expect and the fastest extended progress seen without significant AI help.
  2. It is plausible that the doubling is substantially down to automated research or engineering, rather than more staff, more compute or general productivity.

The word "expect" matters. The line can be met on a forecast, which fits the filing's reference to expectations about future models. The filing's phrase, "dramatically accelerated progress attributable to AI", isn't the exact wording of the policy, which says "dramatic acceleration". Reading them as the same test is the natural reading, since the filing calls it the policy's definition.

"Double the rate" has a specific meaning: "as much progress in one year as one would see in two years at baseline." It compounds, so it is not the same as researchers working twice as fast. The policy says so directly.

Try it: Anthropic's example, extended

81× effective scaleup in one year

The policy's example is a baseline year of 3× compute and 3× algorithmic efficiency, so 9× effective scale. Double the rate gives "something like" 81×. Other rates here are 9 to that power, and the bar is on a log scale. It's arithmetic on the example, not a forecast.

What Anthropic measured

The most recent public measurement is the August 2026 Risk Report. It describes "meaningful acceleration starting in early-to-mid 2025, though by less than a factor of 2", and says the company is fairly confident the 2025 acceleration came from factors other than its use of AI models, while AI has been a key factor in the faster trends continuing since.

Its conclusion: "the overall pace of progress remains below the RSP threshold". It adds that this "is uncertain and subject to some lag." Its leading indicators are left out of the public version as sensitive.

The same report calls the threshold "a potential early warning, rather than evidence that the threat has already materialized." The scenario it treats as most decision-relevant is much further along: compounding automation producing something like a 1,000× to 10 billion× effective scaleup within a year.

What it means, and what it doesn't

It does not say a recursive self-improvement loop is running, that any current Claude model has crossed the line, or that the crossing is certain. The word is "may", and it's about future models.

It does say a top lab now puts this close enough to warn investors, in the same document that describes each version of Claude helping to build the next. Anthropic also says it will release further reports as the policy requires, so the next one is the thing to watch.

Anthropic isn't a neutral narrator here. It is selling shares, and fast progress is part of the pitch, though this sentence appears as a risk. Its numbers are its own, and outsiders can't check most of them.

Reactions

  • Peter Wildeford (@peterwildeford, Nov 2) put the "imminently" sentence next to the one saying Claude won't be used to autonomously build successors, without comment.
  • Andrew Curran (@AndrewCurran_, Nov 2) flagged it as new language on page 29 and spelled out the threshold as one year of progress equal to two at the 2018 to 2024 baseline.

Timeline

Feb 24, 2026
RSP v3.0 frames the line as compressing two years of 2018 to 2024 progress into one.
Jul 8, 2026
RSP v3.4 takes effect, with the current wording.
Aug 2026
Risk Report: below the threshold, uncertain, subject to lag.
Oct 26, 2026
S-1 made public, without the "imminently" sentence.
Nov 2, 2026
S-1/A adds it.

Questions people are asking

Is this the singularity?

No. The threshold is a doubling of the pace, which Anthropic itself calls an early warning. Whether it leads anywhere more dramatic is exactly what's uncertain.

Has Claude crossed the line?

Not by the filing's account: "We have not designated any of our current models as exceeding this threshold."

What is recursive self-improvement?

AI systems doing the work that makes the next AI systems better, so that each round speeds up the next. The filing uses the term, in quotes, and says the policy covers the approach to it.

What happens if Anthropic meets it?

The filing says it is "evaluating the implications" and will release further reports. It also says it is considering policy updates to require human review of certain research decisions and keep humans in control.

Why might a speedup go unseen?

The filing says much of its evidence comes from lagging indicators and that its ability to measure recent acceleration is limited. Benchmarks on released models arrive after the work that produced them.

Updates

November 2, 2026: This page went up. It covers everything known as of November 2, 2026.

Sources