Anthropic's Amendment No. 2 to Form S-1, filed November 10, 2026, states the company made a determination that it has met its Responsible Scaling Policy (RSP) threshold for AI-accelerated research and development — based on expectations for future models and overall progress, not solely on what current deployed models have demonstrated. The same amendment keeps the $70 to $78 IPO range on 825 million shares we covered from Anthropic's roadshow launch.
One week ago, Amendment No. 1 said internal estimates meant the threshold might be met imminently. Amendment 2 upgrades that legal statement to met, while repeating that the evidence mix includes forward-looking capability progress.
Agents that act on your behalf need rules. See how explainx helps secure your agents →
TL;DR
| Question | What the Nov 10 filing says |
|---|---|
| Threshold status? | Met the AI R&D capability threshold under the RSP. |
| Basis? | Expect future models and overall pace — not only observed current models. |
| Still assessing? | Yes — continues to assess underlying data and implications. |
| Business risk? | Safety measures may delay releases or limit availability; may lag competitors. |
| IPO terms? | $70–$78, 825M shares, Nasdaq ANT (unchanged from roadshow post). |
| Wording shift? | Super intelligence / SI used interchangeably with AI per EO 14434. |
The exact risk-factor language
The operative sentence appears in the registration statement's risk factors (SEC filing):
Anthropic writes that its use of AI systems has already been accelerating research and development, that internal estimates indicate it may experience a further increase in the rate of AI capability progress, and that because the RSP threshold applies where Anthropic "observe or expect" a specified degree of accelerated progress, "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."
The filing adds that Anthropic continues to assess the underlying data, is evaluating implications, and will release further reports as its policy requires.
That wording matters for three audiences:
- Regulators and safety researchers tracking RSP versions and METR's independent read.
- Enterprise buyers comparing Anthropic to OpenAI's pacing narrative and Paris Call shutdown pledges.
- IPO investors weighing whether precautionary slowdowns could affect compute commitments already in the tens of billions.
What the RSP threshold is measuring
Anthropic's public Responsible Scaling Policy defines escalating obligations as AI R&D capability grows. The IPO language refers to the threshold where AI contributes enough to R&D that progress roughly doubles how fast frontier capabilities improve — the policy's "dramatically accelerated progress" framing that packed years of historical progress into a single year in earlier definitions.
Crossing the line in the filing does not read as a shutdown trigger. Anthropic describes additional safeguards, ongoing assessment, and future reports — consistent with Dario Amodei's pacing-not-pausing posture ahead of the listing.
The filing also warns about organizational visibility:
As we increasingly leverage Claude to accelerate progress within Anthropic, our organization may lose visibility across certain processes in a way that may become difficult for any individual or group of humans to understand and control.
That sentence sits beside the threshold determination in investor discussions. It is the governance anxiety behind Warren and Levine letters asking for stronger oversight before the IPO.
How METR's estimate fits
On November 5, METR published that Anthropic is likely improving capabilities at about twice its prior-era pace, with perhaps 80% of acceleration attributed to AI-assisted research — while flagging high uncertainty and heavy redactions.
METR also wrote that Anthropic's August position that it remained below the threshold was overstated, and that crossing the early warning line was plausible by mid-July.
Amendment 2 does not cite METR. It advances Anthropic's own RSP determination to met, using expectations language METR skeptics will read as forward-looking rather than retroactive proof on deployed models.
Builders should treat METR's numbers as audit estimates and the S-1 line as legal disclosure — related, not identical.
Precautionary measures and product impact
The same risk factor warns that if capabilities outpace safety work, Anthropic may adopt measures that delay model releases or limit initial availability, citing Mythos through Project Glasswing as a past example.
It explicitly says such measures could slow capability gains versus competitors and hurt revenue growth — unusual candor in a roadshow week also featuring Claude Meetings and the office suite.
For Claude API customers, the practical question is whether new model tiers face staged rollouts or usage caps after this determination. The filing points to that possibility; it does not announce a specific product change on November 10.
Super intelligence wording in the prospectus
Amendment 2 adds definitional language that super intelligence, SI, artificial intelligence, and AI are interchangeable in the prospectus, citing the September 29, 2026 executive order directing federal agencies to prefer super intelligence terminology.
Amendment 2’s Select Defined Terms section spells out the interchangeability; compare S-1/A-2 to the November 2 S-1/A-1. Secondary outlets including Yahoo Finance highlighted the super intelligence shift during IPO week — part of the wider big-tech rebrand pattern explainx.ai tracked after the order.
The wording change does not alter model behavior. It may affect how enterprise procurement templates label vendor risk sections this quarter.
IPO context on the same filing date
Amendment 2 is not only a safety edit. It is the effective roadshow document for:
- 825,000,000 Class A shares at $70.00 to $78.00.
- 801,351,354 primary shares from Anthropic plus selling stockholders' slice (see roadshow launch).
- Cornerstone and cloud spend updates in Amendment 2 commitment tables.
Outlet math near $1.4 trillion fully diluted valuation at the top of the range is media arithmetic, not a line item Anthropic product marketing confirmed.
What builders should do this week
None of this rewrites your Claude integration checklist overnight. It does change the questions worth asking:
| Stakeholder | Question to ask |
|---|---|
| Security | What pause or staged rollout clauses exist if RSP thresholds trigger stronger controls? |
| Legal | Does your contract reference frontier model definitions that shift with Mythos / Fable naming? |
| FinOps | If releases slow, do you have fallback models routed (see frontier comparison hub)? |
| Safety | Do you run your own evals, or rely on vendor cards — Andy Hall-style audit cost applies at scale. |
Anthropic says further reports are coming. Watch the public RSP page and system cards — not rumor threads — for product-facing changes.
Closing the loop in the filing's own words
Amendment 2 spends pages on recursive development — what Anthropic calls closing the loop, where each Claude generation helps train and evaluate the next. The threshold determination sits in that narrative: acceleration is not a hypothetical future, but a process Anthropic says is already speeding R&D and may accelerate further.
The same section warns that lagging indicators may hide acceleration until after it happens. That is why the determination rests partly on expectations for unreleased models. Critics will read that as IPO lawyering; supporters will read it as honest uncertainty about measuring a moving target.
Either way, the disclosure binds Anthropic to follow-on reports under the RSP. The August Risk Report redactions METR criticized become more politically salient now that the company told the SEC the threshold is met, not merely near.
Insurance and tail-risk language investors noticed
The same S-1/A-2 risk factors warn that after a catastrophic AI incident, no single insurer, and potentially the insurance industry in the aggregate, may lack capacity or expertise to underwrite commensurate coverage — language carried from earlier amendments but landing differently beside a met acceleration threshold during IPO roadshow week.
Yahoo Finance relayed Constellation Research’s Ray Wang saying S-1 risks are deliberately overstated — a reminder that legal disclosure and safety policy are different genres. Builders should not treat tail-risk paragraphs as product roadmaps, but enterprise risk officers may still paste them into vendor reviews.
Governance structures named in Amendment 2
The filing repeats Anthropic's public benefit corporation status, Long-Term Benefit Trust with Class T directors, Founder LLC Class F voting block, and Class E for employees. None of that changed on November 10, but the threshold news gives LTBT watchers a concrete trigger: what do Class T directors do when the RSP says acceleration threshold is met?
Anthropic does not answer that in the excerpted risk factors. It says it is evaluating implications and will release reports. If you sell into regulated industries, expect customer security questionnaires to cite this amendment alongside EU probe reporting and Australia mandatory incident rules.
Compare to OpenAI's parallel disclosures
OpenAI's November misalignment cadence — including the July 19 CI report published the same evening — shows a different transparency strategy: incident-sized posts on alignment.openai.com rather than S-1 risk edits.
Anthropic's move compresses safety state into registration statements during IPO week while shipping office collaboration features for enterprise buyers. Builders feel both tracks: product velocity and governance headlines in the same news cycle.
Neither company's disclosure changes your API key today. Together they define what enterprise procurement will ask about in Q4 2026: shutdown time, eval containment, and whether acceleration thresholds trigger slower external releases.
Related reading
- Anthropic IPO roadshow at $70–$78
- Earlier S-1 imminently language
- METR ~2x acceleration estimate
- Amodei pacing-not-pausing
- Paris Call fail-safe shutdowns
Primary sources: Anthropic S-1/A-2 (Nov 10, 2026), Anthropic roadshow announcement, Responsible Scaling Policy, and METR's November 5 assessment.