Anthropic's Claude Solves Two Quantum Gravity Problems

Anthropic says its Claude agents completed a long-unfinished calculation and two theoretical physics problems, though neither result has been peer reviewed.

Updated on Nov. 20, 2026 in Artificial Intelligence

Anthropic's Claude Solves Two Quantum Gravity Problems

Anthropic announced on November 17 that Claude agents solved two theoretical physics problems in quantum gravity, and preprints describing the work were posted on November 7. The agents reported completing a calculation for the Hawking-Page transition at weak coupling, first outlined in 2003 but never finished.

Why it matters

If confirmed, the work would show AI systems contributing original results to fundamental physics rather than only assisting with routine tasks. Because the preprints are primarily AI-generated and have not completed independent peer review, the claims remain unverified by the scientific community.

The agents derived an exact divisor of approximately 9.1266 for the spinless modular bootstrap, which confirms the technique cannot exceed the prior c/8.5 ceiling on the lightest black hole mass. The result is unverified by independent peer review.

The players

Anthropic

Anthropic is an artificial intelligence company that develops the Claude family of AI models and has increasingly promoted their use in scientific research.

The details

Anthropic researcher Hongbin Chen asked Claude to propose open questions, and Claude returned 22 of them, which agents then investigated for about six weeks. For one problem, multiple agents converged on the same result and a physicist using Claude obtained it by a different method, while Anthropic reported formal Lean verification and physicist review for the second.

Timeline

  1. November 7: Preprints describing the results were posted.

  2. November 17: Anthropic announced the results.

The Tech Race

The results extend Anthropic's Claude model family beyond coding and writing into theoretical physics, a direction rival AI developers have also pursued as they compete to show frontier models can perform original scientific reasoning. Whether that positioning holds will depend on how the preprints fare under independent peer review.

For most readers the immediate effect is limited, since the physics results are preprints without peer review. The work does signal that AI tools may increasingly be used in research workflows, which could eventually affect how scientific findings are produced and checked.

The takeaway

Treat AI-generated scientific claims as provisional until independent physicists have reviewed and reproduced them. Readers following this area can watch for peer-review outcomes on the two preprints before drawing conclusions about the reliability of the results.

Further reading

For broader context on how AI systems are being applied to research, see the Artificial Intelligence section.

Source note: This article includes information reported by EWEEK.