Anthropic says its Claude AI models have solved two long-standing open problems in theoretical physics involving black holes and quantum gravity. The findings include completing a complex calculation that physicists had outlined in 2003 but never finished, and deriving an exact mathematical limit for a technique used to investigate possible theories of quantum gravity. The company published its findings on November 17, 2026.
The research demonstrates how advanced AI systems can contribute to scientific investigations that require complex mathematics, independent calculations and collaboration between multiple agents. Anthropic says the models worked on the problems largely autonomously, with physicists checking the results. One of the findings was formally verified using the Lean proof assistant. However, the results do not solve the broader mystery of quantum gravity or establish a complete theory that unifies general relativity and quantum mechanics.
Key Takeaways
- Two physics problems: Anthropic says Claude solved two open problems involving black holes and quantum gravity.
- First breakthrough: The model completed a calculation first proposed in 2003, determining that a black-hole-related phase transition remains abrupt at weak coupling.
- Second breakthrough: Claude derived an exact upper limit in the modular bootstrap method used to constrain possible theories of quantum gravity.
- Fable 5.2 involvement: Claude Fable 5.2 contributed to the modular bootstrap result.
- Multiple AI agents: The research involved several Claude agents working independently and sharing progress.
- Independent checks: Physicists checked the findings against known results, while the second result was formally verified using Lean.
- Important limitation: Neither discovery provides a complete theory of quantum gravity or directly describes our universe.
Claude Solves Two Long-Standing Problems in Black Hole Physics
Anthropic’s research focuses on two mathematical problems connected to the relationship between general relativity and quantum mechanics.
General relativity, developed by Albert Einstein, explains gravity and the behaviour of massive objects such as stars, galaxies and black holes.
Quantum mechanics describes the behaviour of matter and energy at extremely small scales.
Both theories are remarkably successful within their respective domains, but combining them into a consistent theory of quantum gravity remains one of theoretical physics’ biggest challenges.
Black holes are particularly useful for investigating this problem because they involve extremely concentrated gravity and quantum effects.
Anthropic says Claude investigated two questions using a mathematical framework known as holographic duality.
The company describes the findings as contributions to theoretical physics rather than practical discoveries that can immediately be applied to technology.
1. Claude Completes a Black Hole Calculation Left Unfinished Since 2003
The first breakthrough concerns the Hawking-Page phase transition.
In 1983, physicists Stephen Hawking and Don Page demonstrated that a theoretical universe can undergo a sudden transition as its temperature increases.
At lower temperatures, the universe is dominated by a gas of particles. Above a particular temperature, a black hole becomes the dominant configuration.
This behaviour is called a first-order phase transition, similar in principle to water suddenly changing into steam when it boils.
However, the original calculation relied on general relativity and applied to a regime where the theoretical universe is sufficiently large.
Scientists wanted to know whether the same abrupt transition would occur in a much smaller theoretical universe, where classical gravity becomes unreliable.
Why Physicists Could Not Complete the Calculation
In 1999, researchers identified a related phase transition in a quantum theory describing the boundary of this theoretical universe.
But an important question remained unanswered: Was the transition still abrupt, or did it become smooth and gradual?
In 2003, a group of physicists established the calculation needed to settle the question.
The problem involved determining a mathematical quantity that appeared at the three-loop level of the calculation. It required evaluating thousands of complicated terms, making the work extremely difficult to complete manually.
The calculation also involved mathematical infinities that had to be handled carefully to obtain a meaningful result.
Anthropic says Claude completed the calculation by organizing a small research group made up of multiple copies of the model.
The agents worked on different parts of the problem, independently repeating calculations and checking their results against established findings.
What Did Claude Discover?
The model found that the phase transition remains abrupt even in the weak-coupling regime.
This means the transition between the particle-gas state and the black-hole-related state is first-order at both the strong-coupling and weak-coupling extremes examined in the research.
The result suggests that the sudden transition identified by Hawking and Page is not merely a feature of classical general relativity. It also appears in the corresponding quantum theory.
This provides an important connection between the gravitational description of black holes and the mathematical framework used to study quantum systems.
The first research paper, titled The Hawking-Page Transition at Weak Coupling, reports the calculation and its mathematical findings.
2. Claude Finds an Exact Limit in Quantum Gravity Research
The second breakthrough concerns a mathematical technique called the modular bootstrap.
Unlike the first problem, which involved completing an unfinished calculation, this task required finding a new mathematical approach to improve an existing theoretical limit.
The modular bootstrap is used to test whether proposed quantum theories are internally consistent.
Instead of directly constructing a complete theory of quantum gravity, researchers use mathematical constraints to eliminate theories that cannot satisfy the required physical principles.
This approach is particularly useful when scientists do not yet know how to formulate the complete theory they are trying to understand.
What Is the Modular Bootstrap?
The method is used to investigate simplified models of gravity in a universe with two dimensions of space and one dimension of time.
This is commonly described as three-dimensional gravity.
Physicists examine possible quantum theories and determine whether their mathematical properties are compatible with the requirements of gravity and quantum mechanics.
One important question concerns the maximum possible mass of the lightest black hole in such a theoretical universe.
In 2009, physicist Simeon Hellerman introduced a method that established an upper limit for this quantity.
Over the following years, researchers attempted to improve that limit through analytical calculations and numerical experiments.
Numerical results suggested that the limit could be pushed further than the strongest mathematically proven bound, but researchers had not established the exact value.
Claude Identifies a Mathematical Pattern
Anthropic says Claude Fable 5.2 identified a pattern in numerical results that researchers had not previously used to solve the problem.
The model developed a hypothesis involving a mathematical structure called a spectral curve.
Spectral curves are used in several areas of mathematics and theoretical physics, but Anthropic says this approach had not previously been applied to the modular bootstrap problem in this way.
Claude used the approach to derive an exact mathematical expression for the strongest possible limit within the simplest version of the method.
The resulting expression is approximately:
[
\Lambda \approx 9.1266
]
The research establishes that the lightest black hole’s mass must satisfy the corresponding upper bound and proves that this version of the modular bootstrap cannot produce a stronger limit.
The finding is important because it establishes the mathematical limit of a technique that physicists have used for more than 15 years.
However, it does not identify the correct theory of quantum gravity. Instead, it narrows the range of possibilities that researchers need to investigate.
The second paper is titled The Spinless Modular Bootstrap at Large Central Charge.
How Did Claude Work on the Physics Problems?
One of the most interesting aspects of the research is how Anthropic organized the AI investigation.
According to the company, an Anthropic researcher, Hongbin Chen, began exploring Claude’s theoretical physics capabilities as a side project in August 2026.
He asked Claude to identify interesting unsolved problems involving black holes and the bootstrap method.
The model produced a list of 22 candidate problems.
Chen then established separate AI agents to investigate these questions. The agents could share progress updates and collaborate when their research overlapped.
The research continued in the background for approximately six weeks.
During this period, the agents independently explored mathematical calculations, tested hypotheses and compared findings.
The two successful investigations emerged from this broader research effort.
Multiple Agents Independently Checked Their Work
The Hawking-Page calculation involved multiple agents working on separate portions of the problem.
Anthropic says almost every part of the calculation was independently repeated, with results compared against established mathematical findings.
The modular bootstrap breakthrough involved agents working on related mathematical questions.
One agent identified the numerical pattern, while another working on a sphere-packing problem developed a mathematical method that helped produce the final result.
The company says multiple agents independently reached the same conclusion.
This approach is important because complex scientific calculations can contain subtle errors. Independent reproduction provides an additional way to test whether a result is consistent.
Lean Verification Adds Another Layer of Confidence
The second research result was formally verified using Lean in September 2026.
Lean is a proof assistant that allows mathematical arguments to be expressed in a formal language and checked by a computer.
Unlike a conventional written proof, where a reader must examine each step, a formal proof is checked against a defined set of mathematical rules.
This does not automatically establish that the research question is physically relevant or that every modelling assumption is appropriate. It does, however, provide strong evidence that the formalized mathematical argument follows logically from its stated assumptions.
A physicist also checked the main results, according to Anthropic.
The first result was checked against several known special cases, and Anthropic reports that a separate Claude calculation using a different method reached the same answer.
These verification steps are particularly important because AI-generated mathematical research can contain convincing but incorrect arguments.
What Is Holographic Duality and Why Does It Matter?
Both discoveries rely on a mathematical idea called holographic duality.
The concept allows certain theoretical universes containing gravity to be described using a different mathematical theory that does not explicitly contain gravity.
This provides scientists with an alternative way to investigate difficult questions about spacetime and black holes.
The idea is related to the broader holographic principle, which suggests that information about a region of space can be represented through information on its boundary.
For theoretical physicists, this relationship creates a way to translate complicated gravitational questions into problems involving quantum physics.
The two Claude investigations use this framework to study different aspects of black hole formation and quantum gravity.
| Research problem | Main question | Claude’s reported contribution |
|---|---|---|
| Hawking-Page transition | Does the black-hole-related phase transition remain abrupt at weak coupling? | Completed the calculation and found a first-order transition |
| Modular bootstrap | What is the strongest possible limit this method can establish? | Derived an exact upper bound and proved it is optimal within the simplest method |
The distinction is important because the two results address different mathematical challenges.
One completes a calculation that had remained unfinished for decades. The other establishes the precise limit of a mathematical technique.
What These Discoveries Do Not Mean
Despite the significance of the findings, it would be incorrect to say that Claude has solved quantum gravity.
Neither result provides a complete theory unifying Einstein’s general relativity with quantum mechanics.
The findings also do not directly describe the behaviour of black holes in our actual universe.
Instead, they concern specific theoretical models that physicists use to investigate fundamental questions about gravity.
The modular bootstrap result does not identify the correct theory of quantum gravity. It establishes a mathematical restriction within one version of an existing approach.
Similarly, the Hawking-Page result answers a specific question about the behaviour of a theoretical phase transition.
The two papers should therefore be understood as contributions to an ongoing research programme, not as the end of the search for quantum gravity.
How AI Could Change Scientific Research
The findings illustrate how advanced AI systems could increasingly participate in scientific work that involves long calculations, complex mathematical reasoning and multiple stages of verification.
Traditional research can require scientists to spend months or years working through difficult mathematical problems.
AI agents can potentially divide these investigations into smaller tasks, explore alternative approaches and independently check intermediate results.
This could help researchers investigate more questions within the same amount of time.
The implications extend beyond theoretical physics.
Similar approaches could eventually support research in mathematics, materials science, chemistry, engineering and other disciplines where results can be tested using established computational methods.
However, human scientists remain essential for selecting meaningful research questions, evaluating assumptions and determining whether a mathematical result has genuine physical significance.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



