Anthropic says Claude has computed a nine-loop particle-physics amplitude

Anthropic researchers say Claude has completed a nine-loop scattering-amplitude calculation in N=4 super Yang-Mills theory, a formidable theoretical-physics problem that previously sat well beyond the usual chatbot party tricks. The important point is not that Claude produced some impressive-looking algebra. Anthropic says it completed the calculation inside a structured harness using modest computing resources.

Anthropic Watch analysis

What happened

Physicist and science writer Matt von Hippel challenged AI companies to tackle the frontier calculation. Anthropic says its researchers used Claude to execute the work, with the model operating as part of a controlled computational setup rather than simply replying in a chat window. The result concerns a scattering amplitude in N=4 super Yang-Mills theory, a simplified but influential framework used to study difficult questions in quantum field theory.

Anthropic’s account says the calculation reached nine loops, meaning the result incorporates a high-order series of quantum corrections. The company presents it as evidence that language models can contribute to computationally intensive theoretical physics when paired with suitable tools and supervision. The claim is based on Anthropic’s own research report, not an independently reproduced benchmark in the supplied evidence.

Why it matters

This is a more interesting AI result than another leaderboard shuffle because it points towards a workflow: a model helping researchers organise, extend and check a difficult symbolic calculation. If the result survives expert scrutiny, the practical value lies less in replacing physicists than in widening the range of problems they can attack with limited resources.

There is also a useful boundary here. A successful calculation does not by itself show that Claude understands the underlying physics, nor that the method will transfer cleanly to real-world theories. The harness, the checks and the researchers doing the validation matter enormously. The machine may have done the heavy lifting, but nobody should throw away the lab notebook just yet.

Our read

Anthropic has put forward a serious technical demonstration, and it deserves more attention than the usual “AI passes a test” confetti. The real question is whether other physicists can reproduce the calculation, inspect the intermediate steps and use the approach on problems outside this particular theory. A result that can be checked is valuable. A result that can be checked, reused and extended is where the story becomes genuinely consequential.

What to watch

• Whether the full calculation and intermediate checks are independently reproduced.
• Whether researchers apply the workflow to other high-order field-theory problems.
• How much of the work Claude performed versus the structured harness and human researchers.
• Whether similar results can be achieved with smaller or more efficient models. Sources and evidence: Anthropic’s 25 September research report is the basis for the claims about Claude, the nine-loop calculation, Matt von Hippel’s challenge, the structured harness and the reported computing requirements. The supplied evidence does not establish independent reproduction or prove that the model understood the physics rather than successfully executing a constrained workflow.

Discussion spark: If an AI can complete a difficult physics calculation inside a structured, human-designed harness, should we call that scientific discovery, advanced automation, or something in between?

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.