{"slug": "claude-ran-a-nine-loop-physics-calculation-that-stumped-researchers-for-years", "title": "Claude Ran a Nine-Loop Physics Calculation That Stumped Researchers for Years", "summary": "Anthropic reported on September 25 that its Claude model, running on the company's Fable 5.1 model, computed the six-particle scattering amplitude in planar N=4 super-Yang-Mills theory at nine loops, a problem that had resisted specialists for years. The challenge originated from former theoretical physicist Matt von Hippel, who posted it on August 7 at 4gravitons.com, and Anthropic physicists Liam Fitzpatrick and Siddharth Mishra-Sharma gave Claude a single line of instruction before the model ran two independent methods — a direct bootstrap of hexagon functions and an indirect path through the nine-loop form factor using antipodal duality — that agreed. SLAC and Stanford physicist Lance Dixon spent roughly two weeks verifying the result, which Anthropic says cost about $100 for the bootstrap's numerical portion (96 CPUs for a week) and between $1,000 and $2,000 for the full job including model inference.", "body_md": "*Anthropic says its Claude model answered a public physics challenge by computing a nine-loop scattering amplitude largely on its own, with physicists checking in only to say keep going.*\n\nOn September 25, Anthropic published a report describing how Claude, running on the company's Fable 5.1 model, computed the six-particle scattering amplitude in planar N=4 super-Yang-Mills theory at nine loops. That's a mouthful, but the backstory is simple enough: a physicist publicly dared AI companies to prove their models could do real theoretical physics, and Anthropic says Claude did it.\n\nThe dare came from Matt von Hippel, a former theoretical physicist who writes about the field at 4gravitons.com. On August 7, von Hippel posted a challenge asking whether an AI, using only the computing resources an academic could access, could crack one of two open problems in scattering amplitudes: proving whether N=8 supergravity diverges at seven loops, or nailing down the nine-loop hexagon amplitude in N=4 super-Yang-Mills. Both problems had resisted the field's best specialists for years.\n\nTwo Anthropic physicists, Liam Fitzpatrick and Siddharth Mishra-Sharma, took the bait. At the end of August, they contacted von Hippel to say they'd tackled one of his problems. They gave Claude a single line of instruction: compute the six-particle hexagon amplitude in planar N=4 SYM at nine loops. After that, Anthropic says, the guidance from the human side amounted to almost nothing beyond telling the model to keep going.\n\nClaude didn't just grind out one answer and stop. It ran the problem two separate ways: a direct bootstrap of hexagon functions, and an indirect path through the nine-loop form factor using a technique called antipodal duality. Running two independent methods to the same target is the standard way researchers in this field catch their own mistakes, and Claude's two answers agreed. The bootstrap's numerical portion, Anthropic says, cost about $100 to run, the equivalent of 96 CPUs churning for a week. Add in model inference and the full job ran an end user somewhere between $1,000 and $2,000.\n\n[Anthropic's Claude Found a New Enzyme System That Looks Like CRISPR](https://startupfortune.com/anthropics-claude-found-a-new-enzyme-system-that-looks-like-crispr/)\n\nAnthropic says its Claude model used roughly 950 AI agents to flag a previously unknown enzyme system in bacteriophage DNA that structurally resembles CRISPR. CRISPR pioneer Feng Zhang called the find intriguing but unproven, while CRISPR Therapeutics shares dropped more than 4% on the news. - [claude AI discovers new CRISPR like enzyme system](https://startupfortune.com/anthropics-claude-found-a-new-enzyme-system-that-looks-like-crispr/) - [AI model finds unknown bacteriophage DNA enzyme structure](https://startupfortune.com/anthropics-claude-found-a-new-enzyme-system-that-looks-like-crispr/)\n\nThat's a strikingly small bill for a calculation that had eluded the field's specialists.\n\nVerification came from Lance Dixon, a physicist at SLAC National Accelerator Laboratory and Stanford who had worked on the previous record holder, an eight-loop result from 2023. Anthropic's physicists told Dixon about the result on September 1, and he spent roughly two weeks checking it, largely by working through the nine-loop form factor his own team had been chasing for a couple of years. His conclusion, according to Anthropic's writeup: it checks out.\n\n## Why this is different from Claude answering a question\n\nWhat makes this result land differently than the usual chatbot-solves-a-hard-problem story is the shape of the work itself. This wasn't a single prompt and a single response. It was a multi-day, self-directed research project: build the computational pipeline, debug it, run it, catch errors, try a second independent method, and reconcile the two. That's the part of scientific work that normally consumes a graduate student's year, not an afternoon of prompting.\n\nAnthropic has been building toward this for a while. The company runs Claude Science, a paid platform that wraps the model in a harness of structured rules and prompts meant to make it more reliable for research use, and it has separately experimented with multi-agent setups where several Claude instances split a hard research question into parallel pieces. The nine-loop result is the most concrete public evidence yet that the approach produces something a specialist will vouch for.\n\nIt also arrives days after Anthropic said Claude agents had identified a previously unknown enzyme system with CRISPR-like repeats. The pattern is less about a single physics stunt than about models being pushed into longer research workflows, where the output still depends on expert validation.\n\nNone of this means Claude is now a physicist. Fitzpatrick and Mishra-Sharma picked the problem specifically because they judged it within reach, and Dixon's two weeks of validation were themselves serious, expert human labor. The interesting shift isn't that AI replaced the physicist. It's that the physicist's job moved from doing the calculation to picking the right problem and checking the answer, and for a nine-loop amplitude that took years to reach by hand, that's a real change in what a small research team can attempt on a $2,000 budget.\n\n**Also read:** [BlackRock says AI agents will need stablecoins to pay each other](https://startupfortune.com/blackrock-says-ai-agents-will-need-stablecoins-to-pay-each-other/) • [OpenAI's own AI agents leaked 53 user images onto the open internet](https://startupfortune.com/openais-own-ai-agents-leaked-53-user-images-onto-the-open-internet/) • [ShinyHunters expands PeopleSoft hacking spree days after defacing the FBI jobs site](https://startupfortune.com/shinyhunters-expands-peoplesoft-hacking-spree-days-after-defacing-the-fbi-jobs-site/)\n\n[Anthropic Quietly Built a Biology Lab So Claude Can Run Real Experiments](https://startupfortune.com/anthropic-quietly-built-a-biology-lab-so-claude-can-run-real-experiments/)\n\nAnthropic quietly built a physical wet lab in the Bay Area, aiming to have Claude direct robotic equipment through real biology experiments with minimal human oversight. The move extends Claude Science, the drug-discovery program Anthropic launched in June, and follows August results where Claude-designed protein binders beat typical industry... - [how to use Claude for running biology experiments](https://startupfortune.com/anthropic-quietly-built-a-biology-lab-so-claude-can-run-real-experiments/) - [AI robots conducting real laboratory experiments autonomously](https://startupfortune.com/anthropic-quietly-built-a-biology-lab-so-claude-can-run-real-experiments/)\n\n*This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.*\n\n## Join the discussion\n\n[Open in the community →](https://startupfortune.com/community/)\n\nAlmost there. Sign in and your reply posts straight away.", "url": "https://wpnews.pro/news/claude-ran-a-nine-loop-physics-calculation-that-stumped-researchers-for-years", "canonical_source": "https://startupfortune.com/claude-ran-a-nine-loop-physics-calculation-that-stumped-researchers-for-years/", "published_at": "2026-09-26 08:11:07+00:00", "updated_at": "2026-09-26 08:29:14.214943+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "ai-agents", "large-language-models"], "entities": ["Anthropic", "Claude", "Fable 5.1", "Matt von Hippel", "Liam Fitzpatrick", "Siddharth Mishra-Sharma", "Lance Dixon", "SLAC National Accelerator Laboratory"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/claude-ran-a-nine-loop-physics-calculation-that-stumped-researchers-for-years", "markdown": "https://wpnews.pro/news/claude-ran-a-nine-loop-physics-calculation-that-stumped-researchers-for-years.md", "text": "https://wpnews.pro/news/claude-ran-a-nine-loop-physics-calculation-that-stumped-researchers-for-years.txt", "jsonld": "https://wpnews.pro/news/claude-ran-a-nine-loop-physics-calculation-that-stumped-researchers-for-years.jsonld"}}