Anthropic says Claude solved two open problems in black-hole physics; no outside check yet
Both papers were written mostly by AI and checked by one unnamed physicist. Neither has been peer reviewed.
On 9 November 2026, in the launch post for Claude Fable 5.2, Anthropic said the model had solved two open problems in theoretical physics. Both concern black holes and use holographic duality: the idea that some universes with gravity can be described exactly by a quantum theory. Anthropic posted two preprints (papers not yet reviewed by a journal), dated 7 November. They list Anthropic as contributor and name no authors.
The first asks how black holes form as such a universe heats up. In 1983 the physicists Hawking and Page showed they appear abruptly, like water boiling, but their method worked only for large universes. A calculation for small ones was set out in 2003 and never finished. The 42-page preprint completes it and finds the change is abrupt there too.
The second concerns a method physicists have used for more than 15 years to rule out candidate theories of quantum gravity in a simplified universe. A key number in it was known only to lie between 8.503 and 12. The 131-page preprint says it is exactly about 9.1266.
On the first problem, a physicist prompted the model and checked the answer against known special cases. On the second, the preprint says the model reached the main result on its own, the public Claude Opus 5.5 later reached it too, and a physicist checked every main result.
What it means
If the results hold, an AI system did original work in theoretical physics, once with little human steering. Physicists outside the company have not yet said whether the results are right or how much they matter.
Caveats
- These are Anthropic’s own claims. Neither preprint is peer reviewed, and we found no named outside physicist confirming or disputing them.
- Both preprint pages are labelled “Primarily AI-generated text”, with only some parts marked as understood by a human.
- The launch post credits Claude Fable 5.2. The preprints credit an internal Claude research model, so which model did the work is not fully clear.
- The preprint pages mention no computer-checked proof, unlike several recent AI maths results. The first result rests on spot checks and a second Claude calculation that is not included.
- Anthropic does not say how many other problems were tried without success.