Anthropic says autonomous agents running its Claude Fable 5.2 model have produced results on two open problems in theoretical physics related to black holes. The company published the work on November 17 alongside two papers, and says a physicist checked every main result.
The first result concerns the Hawking-Page phase transition in N=4 super Yang-Mills theory at weak coupling. According to Anthropic, Claude completed a three-loop calculation that physicists set up in 2003 but never finished, and found that the transition is abrupt at weak coupling, matching the abrupt transition Hawking and Page found at strong coupling. Anthropic notes that this makes it natural to expect the transition stays abrupt at intermediate coupling, but that this has not been proven.
The second result is in the modular bootstrap approach to 3D quantum gravity. Claude found an exact upper bound, M1 ≤ c/Λ with Λ ≈ 9.1266, and showed it is optimal for this version of the method, using a spectral-curve technique not previously applied to the problem. Anthropic says the bound does not reach c/12, the mass at which black holes first appear, and does not by itself pin down a theory of 3D quantum gravity.
The project was set up by Anthropic researcher Hongbin Chen, who asked Claude to list open problems in black hole physics and bootstrap methods; Claude proposed 22 candidates. The agents worked in the background for about six weeks. Anthropic says the phase-transition calculation was run twice by independent agents and compared against published results, and two other agents independently reached the bootstrap answer. A Claude agent had earlier failed on the bootstrap problem across 61 parallel attempts.
Anthropic stresses that neither result describes our own universe directly or has practical applications. The work adds to a growing set of cases where long-running AI agents are used for multi-week research calculations, with human experts verifying the output rather than producing it.