Anthropic says its Claude model has solved two open problems in theoretical physics that had stumped researchers for years. The findings concern black holes and the search for a theory of quantum gravity, and one result was formally verified by computer.
A calculation left unfinished since 2003
The first problem concerns what happens when a theoretical universe heats up. At low temperatures it fills with gas, but above a certain point a black hole forms instead. In 1983, Stephen Hawking and Don Page showed that switch is abrupt, like water suddenly boiling.
That calculation relied on general relativity, which holds only when the universe is vast. As the model universe shrinks toward the tiny scale where spacetime loses its shape, the gravity side of the problem breaks down, and only the quantum side can be calculated at all.
In 1999, physicists found that a phase transition exists in that small-scale regime too, but nobody knew whether it was abrupt or smooth. In 2003, a group of physicists set out the calculation that would settle it, involving thousands of terms and infinities that had to be tamed carefully.
Claude finished the job by running a small research group of copies of itself, each working out almost every piece of the calculation independently and checking the math against known results. It showed the transition is abrupt, suggesting black-hole formation is built into the underlying quantum theory itself.
An exact ceiling physicists had only guessed
The second problem uses a method called bootstrapping, which rules out wrong theories of quantum gravity by setting a ceiling on how heavy the lightest black hole can be. Since 2009, physicists have lowered that ceiling step by step, guided by numerical hints they could not prove.
Claude Fable 5.2 spotted a pattern in the numerical results that physicists had missed, then borrowed a method from elsewhere in mathematics and physics. It calculated the precise ceiling for the simplest version of the technique and proved it could not be lowered.
The method itself came from an unexpected direction. Another Claude agent, working on how tightly spheres can be packed into high-dimensional space, turned the pattern into a usable method. Two more agents independently reached the same answer, and in September the results were verified in Lean, a proof checker.
Why It Matters
Neither result describes our own universe directly, and neither has a practical application. Their value is different: they show an AI system doing original theoretical work, not just answering questions, and doing it well enough to pass the field’s toughest checks.
The work also hints at how research itself may change. The project began in August as a side experiment, when Anthropic researcher Hongbin Chen asked Claude to list open problems in black-hole physics. The agents then ran in the background for about six weeks, with no special setup.
Anthropic notes that a physicist working independently reached the same Hawking-Page answer by a different method, a useful cross-check. Claude then wrote the paper with minimal guidance, a physicist checked every main result, and the preprints are now available for other researchers to examine.
Background
Claude Fable 5.2 arrived in early November as Anthropic’s flagship model, positioned against OpenAI’s latest systems. It has already shown strong results on freelance work, completing 41 percent of real freelance jobs in one benchmark, and the physics results add research to its record.
The findings do not identify the correct theory of quantum gravity, and Anthropic is careful to say so. They narrow the search instead, pointing physicists toward the theories that remain possible and showing what multi-agent AI research can now do.
Source: Anthropic research post


0 Comments