AI & Innovation

Claude Fable 5.2 solves two black hole physics problems, Anthropic says

By Anika SharmaAI & Innovations Editor, Peepals Tech
WhatsAppX
Representative image: galaxy stars night sky

Representative image: r-q / Pixabay

Anthropic said on 17 November that Claude Fable 5.2 agents, working for six weeks, solved two of 22 open problems in black hole physics, checked by a physicist.

What Does Anthropic DO: what happened

Anthropic, the company that makes the Claude family of artificial intelligence (AI) models, said on Tuesday, 17 November 2026, that its Claude Fable 5.2 model had solved two open problems in the theoretical physics of black holes. According to the company's research post, Claude was asked to list unsolved problems in the field, proposed 22, and separate AI "agents" then worked on them on their own for about six weeks. Anthropic says two of the problems were solved, with the results checked by a physicist and parts of the work verified in Lean, a computer language for checking mathematical proofs.

What Does Anthropic DO: key facts

  • Announced: Tuesday, 17 November 2026, in a research post on Anthropic's website.
  • Model: Claude Fable 5.2, according to Anthropic.
  • How it worked: researcher Hongbin Chen asked Claude for open problems on black holes and the "bootstrap" method; Claude proposed 22 and one agent was assigned to each.
  • Time: the agents ran in the background for about six weeks.
  • Result 1: the Hawking-Page transition, the point at which a hot gas in a certain theoretical setting turns into a black hole, stays abrupt even at weak coupling, completing a calculation that had been set up years earlier but never finished, according to summaries of the work.
  • Result 2: an exact ceiling, about c/9.1266, on the mass of the lightest black hole that the spinless modular bootstrap method can guarantee, shown to be the best that method can give.
  • Limits: the bootstrap result still cannot reach c/12, the level at which black holes are expected to first appear, so the larger question stays open.

What Does Anthropic DO: why it matters

The question behind this announcement is whether AI models can do original research rather than only summarise existing work. Anthropic's post describes an AI system choosing the problems itself, working on them for weeks without supervision and producing results that a human expert then checked. That is a different claim from a model answering a question it is given.

Two terms help. "Coupling" is a measure of how strongly the particles in a theory interact; physicists often know what happens when interactions are strong or weak but not both. The Hawking-Page transition was known to be sharp at strong coupling; the new work says it remains sharp at weak coupling too. The "bootstrap" is a method that rules out impossible answers using only basic consistency rules, without solving the full theory. It gives upper limits, and the new result finds the tightest limit that one version of the method can reach.

The numbers also show the limits. Only two of 22 attempted problems were solved, and the second result is an improvement within a method rather than an answer to the deeper question of when black holes appear. These are Anthropic's claims about its own model, published by the company, and the wider physics community will judge the papers over time.

For Indian readers, the work is a marker of where the frontier is moving: from chatbots that answer questions to agents that run long research projects. For Indian universities and research institutes, the practical question is how such tools should be checked and credited before their results are trusted.

Peepals calculation: two solved out of 22 attempted problems is a success rate of about 9% (2 ÷ 22 ≈ 0.091). On the bootstrap result, a ceiling of c/9.1266 is about 31% higher than c/12 (12 ÷ 9.1266 ≈ 1.31), which shows how far the method still is from the level where black holes are expected to appear. Our calculation from the reported figures.

ItemFigure
Open problems proposed by Claude22
Problems solved, per Anthropic2
Time the agents ranabout 6 weeks
Bootstrap ceiling foundabout c/9.1266
Level still out of reachc/12

What Does Anthropic DO: impact

Short-term: The announcement will add to debate on how far AI can go in science, and on how its results should be checked and credited. Physicists will examine the two papers on their merits.

Long-term: If AI agents can reliably work through open problems over weeks, research groups could use them to explore many questions at once, with humans choosing and checking. That would change how theoretical research is done, but only if verification keeps pace.

Who is affected: Theoretical physicists and mathematicians, AI research labs and their competitors, universities that set rules on AI use in research, and students who will study in fields where such tools are common.

What Does Anthropic DO: key takeaway

Treat this as a company's own claim of a meaningful but narrow advance, and watch how independent physicists respond to the two papers before drawing bigger conclusions about AI doing science.

What Does Anthropic DO: questions and answers

What does Anthropic do?

Anthropic is an AI company that builds the Claude family of AI models, used as chat assistants and as "agents" that can carry out long tasks. On 17 November 2026 it said its Claude Fable 5.2 model had solved two open problems in black hole physics.

What is the Hawking-Page transition?

It is the change, in a theoretical model, from a hot gas to a black hole as temperature rises. Anthropic says Claude showed that this change stays abrupt even when the particles in the theory interact weakly.

How were Claude's results checked?

According to Anthropic, the main results were checked by a physicist and parts of the work were verified in Lean, a programming language used to check mathematical proofs by computer.

Does this mean AI can now do physics research on its own?

Not entirely. A human researcher started the project, only two of 22 attempted problems were solved, and the results were checked by people. The work shows AI agents contributing to research, not replacing researchers.

Disclaimer: Prepared by the Peepals newsroom from publicly available sources with AI assistance. Information is accurate to the best of our knowledge at the time of publication and may change. Images may be representative. Not professional advice. Report an error via our contact page.

#what does anthropic do#Anthropic#Claude Fable 5.2#black hole physics#AI research agents#Hawking-Page transition

By Anika Sharma, AI & Innovations EditorCovers AI & Innovation and more for Peepals Tech. All stories by Anika Sharma

Your week ahead, every Sunday. Free.

Exam and result dates, IPOs opening, scheme deadlines, government holidays and the RBI decision for the next 7 days, plus the week's top stories. One email a week, unsubscribe in one click.

Anika Sharma is a pen name of the Peepals Global Editorial Team, which researches, fact-checks and publishes every story. The portrait is an illustration.

Responsible for this story: the Peepals Global Editorial Team, led by editor Prashant Sonar. How we report

More from Peepals Tech

Your week ahead, every Sunday. Free.

Exam and result dates, IPOs opening, scheme deadlines, government holidays and the RBI decision for the next 7 days, plus the week's top stories. One email a week, unsubscribe in one click.