Anthropic says Claude Fable 5.2 solved two open problems in black hole physics
Anthropic says Claude Fable 5.2 solved two open theoretical physics problems on black holes and quantum gravity, one a long-unfinished 2003 calculation.
Anthropic said on November 17 that Claude Fable 5.2 solved two open problems in theoretical physics about black holes and quantum gravity. The first completes a calculation on a black hole phase transition that physicists set up in 2003 and never finished. The company says the calculation shows the transition is abrupt at both weak and strong coupling. The second pins down an exact ceiling on the mass of the lightest black hole in a simplified model of three-dimensional quantum gravity. Anthropic states that neither result has direct practical applications.
The first problem goes back to 1983, when Stephen Hawking and Don Page showed that in anti-de Sitter space a hot universe switches abruptly from a gas of particles to a black hole, much as water suddenly boils. Their calculation relies on general relativity, which only holds when the universe is large compared with the tiny scale at which that description breaks down. Holographic duality offers a second route. The same universe can be described by an ordinary quantum theory on its boundary, with no gravity at all. That quantum description can in principle be calculated even where the gravity description breaks down. In 1999 Bo Sundborg, and later Ofer Aharony and colleagues, showed that this quantum theory, N=4 super Yang-Mills, also undergoes a transition at weak coupling. It was unknown, however, whether that transition was abrupt or gradual.
The question hinged on a single number that first appears three loops into the perturbative expansion. Anthropic's post says computing it involves many thousands of terms, called diagrams, which is far too many to do by hand. In 2005 the same group finished the calculation for a simpler cousin of the theory and found an abrupt transition. The version for N=4 super Yang-Mills, more than 10 times larger, was never completed. There was also a conceptual difficulty. Adding up the parts naively gives an infinite answer that has to be tamed before a meaningful result emerges.
According to the post, Anthropic researcher Hongbin Chen asked Claude in August to list open problems in theoretical physics focused on black holes and the bootstrap method. Claude returned 22 candidates, and Chen set up a separate agent on each one. The agents ran in the background for about six weeks. For the Hawking-Page calculation, almost every piece was computed twice by agents working independently, and parts were compared against published results. The post says a physicist who had independently used Claude on the same problem later shared a result. A colleague found that Chen's agent had reached the same answer earlier, by a different method. The Hexagon preprint describes the result differently. An internal Claude research model, prompted by a physicist, obtained it, and the physicist checked it against several known special cases.
The second problem concerns 3D gravity, a model with two spatial dimensions that still contains black holes. Physicists cannot yet write down a complete theory of it, so they use the bootstrap, which rules out candidate theories by showing that they contradict themselves. Simeon Hellerman introduced the modular bootstrap in 2009. In his original paper he showed that the mass of the lightest black hole, M1, can be no larger than c/6, where c is the central charge, a count of the microscopic building blocks. Later work pushed the ceiling down. Hartman, Mazáč and Rastelli proved a ceiling of c/8.5 analytically, while numerical calculations hinted at roughly c/9.1. They could not prove a ceiling that low.
Claude's contribution, as the post describes it, was to notice a pattern in the numerical results, in which the ceiling shifted slightly as c grew. Using a mathematical tool called a spectral curve, which is well known elsewhere in mathematics and physics but had not been applied to this problem, Claude derived what the post calls an exact ceiling of about c/9.1266. The preprint gives the constant as 12π² divided by (π² + 4 arcsinh² 1). It says no linear functional in a broad class can improve on it by more than a small correction. It also gives an explicit functional that yields a bound slightly above that constant, with a correction of 0.483c^(1/3). The coefficient of the c^(1/3) correction to the optimal bound is only a conjecture, supported by numerics up to c=2048.
The route to the answer was indirect. One agent, working on a related problem about packing spheres in high-dimensional space, turned the pattern into a method. It found that the method reproduced a sphere-packing result that an OpenAI model had recently proved. Claude then applied the same method to the bootstrap problem, which an earlier Claude agent had failed to solve in 61 parallel attempts. Two other agents reportedly reached the same answer independently. Half of the final proof, the part showing the ceiling cannot be lowered, follows the OpenAI strategy. The results were checked in the Lean proof assistant in September. Claude wrote the paper with minimal guidance, and a physicist checked every main result.
The model naming is inconsistent across the materials. The blog credits Claude Fable 5.2 throughout. The Hexagon preprint says the main result was first obtained by an internal Claude model, and that the publicly available Claude Opus 5.5 later reached the same result independently. The post doesn't explain the discrepancy.
The limits the post itself acknowledges are specific. The modular bootstrap method, as used here, cannot on its own push the ceiling down to c/12, the value at which black holes first appear, so that would require other ingredients. The post also says the work doesn't identify the correct theory of 3D quantum gravity, though it may point physicists in the right direction. On the Hawking-Page side, the post's suggestion that the transition stays abrupt at every coupling in between is described as the most natural possibility, not something the calculation establishes.
What changed
Anthropic says Claude Fable 5.2 solved two open theoretical physics problems on black holes and quantum gravity, one completing a 2003 calculation.
Type: Research · Applies to: Claude Fable 5.2
Why it matters
The claim is that long-running, largely autonomous agents produced checked research results in a hard field. Teams planning agent-driven R&D may find that a useful signal. The results are company-reported, and the company itself says they do not settle the underlying question.
Original source
Where we saw it
Anthropic Research
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works