On this page
Anthropic said in a research post dated 17 November 2026 that its Claude model has "solved two open problems in theoretical physics", both linked to black holes and quantum gravity. The company published two preprints on the Hexagon server, deposited on 7 November 2026, to back the claim. At the time of writing we found no independent physicist assessment of either result, so everything below is Anthropic's account unless stated otherwise.
What exactly is Anthropic claiming?
Anthropic's research post names the model as Claude Fable 5.2 and describes two results. Its earlier model announcement of 9 November 2026 listed the same work among prerelease achievements, and states that Fable 5.2 is available on all platforms.
| Problem 1 | Problem 2 | |
|---|---|---|
| Topic | Hawking-Page transition at weak coupling | Spinless modular bootstrap at large central charge |
| Question | Is the switch from a hot gas to a black hole abrupt in weakly coupled N=4 super Yang-Mills? | How low can the bound on the lightest black hole in 3D gravity go? |
| Anthropic's claim | The transition is abrupt, as it is at strong coupling | An exact ceiling, with a constant of about 9.1266, and the method cannot do better |
| Preprint | 2611.00175v1 | 2611.00174v1 |
The Hawking-Page problem
Hawking and Page showed in 1983 that, in a particular kind of universe, a hot gas of particles gives way abruptly to a black hole. A dual description using a quantum theory without gravity exists, and Anthropic says a calculation set up in 2003 to test the weak-coupling version was never finished because it was too laborious.
According to the preprint, the answer hinges on the sign of a coefficient in the effective action. The paper computes it and concludes the transition is first order, meaning abrupt. Anthropic's post is careful on one point: that the transition stays abrupt at every coupling in between is described only as the most natural possibility, not as something proven.
The modular bootstrap problem
The second preprint studies a technique used for years to constrain candidate theories of quantum gravity in a simplified universe. It states an exact value for the best possible slope, about 9.1266, and says this is the limit of this simple version of the method. Anthropic says the proof was checked in the Lean proof assistant in September 2026.
The result does not identify the correct theory of 3D quantum gravity, and it does not reach the c/12 threshold at which black holes first appear. Neither result, Anthropic says, describes our own universe directly.
Who checked the work?
Here the record is thin. Anthropic says a physicist "checked every main result", but the post does not name that person. The Hawking-Page preprint says a physicist prompted an internal Claude research model and checked the output against several known special cases. A separate Claude computation using a different method, not included in the paper, reportedly gave the same answer.
Both Hexagon pages label the text "Primarily AI-generated" and give no authors beyond "Anthropic". Neither page states that the work has been peer reviewed. Anthropic's announcement carries a footnote saying external evaluators did not review the post and may not agree with its claims.
Which model did the work?
Accounts differ, and readers should treat the attribution with care. Anthropic's post credits Claude Fable 5.2. The bootstrap preprint says an internal Claude research model first obtained the result autonomously, and that the publicly available Claude Opus 5.5 later reached it independently. The Hawking-Page preprint also refers to an internal research model. We could not establish from the sources how these descriptions line up.
What has anyone outside Anthropic said?
Coverage so far largely repeats Anthropic's account. Crypto Briefing, writing on 9 November 2026, said the findings are company-reported and that their value depends on whether the physics community confirms them. It quoted no physicists. An independent write-up by Pasquale Pillitteri, dated 20 November 2026, likewise found no independent expert comment on these two results and noted the inconsistent model attribution. These are secondary sources, not expert verification.
What is not independently verified?
- That the Hawking-Page transition is abrupt at weak coupling, which rests on a multi-step calculation checked by an unnamed physicist and a second Claude computation.
- That the bootstrap ceiling is optimal for the simple method, though the proof is said to have been formally checked in Lean.
- The claim that Claude did the work largely autonomously, which comes from Anthropic's own description.
- Which specific model produced each result.
Analysis: why this matters, and why caution is sensible
This is our own reading, not a finding from the sources. The two results are narrow. They answer technical questions in idealised settings, and Anthropic itself says they do not describe our universe. The more significant claim is about process: a model proposing problems, calculating, proving and writing up results with limited human guidance.
That is also what makes independent review important. A preprint labelled mostly AI-generated, with no named human authors, is hard for the field to assess until named physicists engage with it. The Lean verification is a strong check on one result, but it checks that a proof is valid, not that the question matters. We would expect clearer signals from the field to emerge over the coming weeks as researchers read the papers.
For context on how AI-assisted verification has been reported elsewhere, see our piece on DeepMind's formal verification of the Signal protocol, and for the company's wider position, our explainer on the Anthropic IPO filing.
Sources
- Anthropic: Claude Fable 5.2 solves two open problems in the physics of black holes
- Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2
- Preprint: The Hawking-Page transition at weak coupling (Hexagon 2611.00175v1)
- Preprint: The spinless modular bootstrap at large central charge (Hexagon 2611.00174v1)
- Crypto Briefing coverage, 9 November 2026
- Pasquale Pillitteri: Claude solves two black hole problems, and what is actually proven
FAQ
Frequently asked questions
That is Anthropic's claim, published on 17 November 2026 with two preprints. No independent physicist assessment of the results had been found at the time of writing, and neither preprint states that it has been peer reviewed.
The first asks whether the Hawking-Page transition, the switch from a hot gas to a black hole, is abrupt in weakly coupled N=4 super Yang-Mills theory. The second asks how tight a bound can be placed on the lightest black hole in 3D gravity using the spinless modular bootstrap.
Not according to the Hexagon pages, which do not state any peer review. They are version 1 preprints deposited on 7 November 2026 and labelled as primarily AI-generated text.
Anthropic says neither result has practical applications or describes our universe directly. Both concern simplified or idealised settings used to study quantum gravity.
Spotted an error? Tell us. Read how we report and correct stories in our editorial policy.
Written by
Mainstream Tech
Newsroom
News and analysis on technology, AI, business, politics and international affairs, from our newsroom in London.
More from Mainstream



