Claude solves two open problems in the physics of black holes

Anthropic reports that Claude worked through two open problems in theoretical physics, both tied to black holes and the search for a quantum theory of gravity. The effort began in August 2026 as a side project of Anthropic researcher Hongbin Chen, who asked Claude for a list of unsolved questions on black holes and the bootstrap method. Claude produced 22 candidate problems, and the company set up a separate agent for each, running them for about six weeks.
The first result concerns the Hawking-Page transition, the temperature at which a hot gas turns into a black hole. Hawking and Page described that switch as abrupt in 1983. A 2003 follow-up calculation was never finished, and its answer for weakly coupled systems hinged on one number that appears three loops into the work. Claude completed the calculation and found the switch stays abrupt.
The second result concerns the modular bootstrap, a method used since 2009 to bound the lightest black hole in a three-dimensional model of quantum gravity. Claude identified the spectral curve that governs the method at large central charge and used it to show the lightest black hole weighs no more than c/Λ, where Λ is about 9.1266. Earlier work had bracketed that constant between 8.503 and 12. Anthropic says a physicist checked the main results, and the bootstrap proof was verified in Lean, a formal proof checker. Both papers are preprints that have not been peer reviewed.
Why it matters
For companies selling AI models, this is a public test of long-running agent research: 22 candidate problems, about six weeks, and two results. Buyers in science and R&D will likely ask how the work was checked, how many attempts failed, and who answers for a claim when the preprint lists Claude as the sole author.
For investors and operators, the useful signal is the workflow: parallel agents, repeated runs, formal checks, and a human reviewer at the end. Those steps can be priced, audited and sold. The same pattern means that claims of solved research will be judged on reproduction and peer review, so teams publishing such claims should plan for that scrutiny before they announce.
Key points
- 1Anthropic says a 2003 Hawking-Page calculation was finished and the transition stays abrupt at weak coupling.
- 2The modular bootstrap result caps the lightest 3D quantum gravity black hole at c/Λ, with Λ about 9.1266.
- 3Earlier work had only bracketed the constant Λ between 8.503 and 12, according to commentary on the paper.
- 4Anthropic says 22 candidate problems were assigned to separate agents, and two produced results.
- 5Both preprints await peer review, and the bootstrap proof was formally verified in Lean.