Physicist Nathaniel Craig has described how GPT-6 Astra helped him develop a candidate lattice definition of the Standard Model of particle physics. In a 15 November guest post on OpenAI’s science blog, the first on that blog, Craig wrote that Astra “took the lead in developing the detailed proof.” The result, he said, is “a candidate nonperturbative lattice definition for the Standard Model, potentially answering a question that had been open for more than fifty years.”
The underlying paper is a preprint, not a peer-reviewed journal article. Craig’s October paper, arXiv:2610.09093, is titled “A gauge-invariant measure for lattice chiral fermions” and lists affiliations at the University of California, Santa Barbara, its Kavli Institute for Theoretical Physics, and OpenAI. The abstract says the work constructs a gauge-invariant measure for overlap fermions on four-dimensional Euclidean lattices, in a class of nonabelian chiral gauge theories that includes the Standard Model, and concludes: “This provides a candidate nonperturbative lattice definition of the Standard Model.”
In the blog, Craig sets out why that definition has been hard to give. The Standard Model, he wrote, agrees with experiment across a huge energy range, yet much of its richness sits outside perturbation theory, the method of treating interactions as small corrections. A lattice — a grid representing spacetime — is a way to define the theory without that crutch, and to put it on a computer. The difficulty, which the preprint also describes, is accommodating the Standard Model’s chiral fermions without sacrificing gauge invariance.
Craig calls the result a beginning, and one question remains open
Craig leads the particle theory group at UCSB and is a permanent member of KITP; the OpenAI post also describes him as a visiting scientist at the company. He wrote that he already had a strategy for the problem, but “had not been able to prove that it worked in all the cases it needed to cover.” Astra, he said, “helped turn that strategy into a construction and took the lead in developing the detailed proof,” including how to carry force fields onto finer grids and show that comparisons on those grids would converge.
He also recorded help from human experts. Lattice field theorists Maarten Golterman, Andreas Kronfeld and Yigal Shamir, he wrote, “graciously spent time looking over Astra’s proof and asking probing questions that helped me to build confidence in the result.” That is a reading by named specialists, not a journal referee report.
Craig’s own caveats are part of the claim, not a footnote. The construction, he said, “is much more a beginning than an end.” “Scientific dialogue among experts is required to decide if this prescription has all the properties we desire.” The method “is far more an answer in principle than an answer in practice; much work lies ahead to make it useful for realistic lattice simulations.” He added that he is “thinking a lot about whether this construction satisfies reflection positivity at finite lattice spacing,” a condition he described as helping to connect lattice calculations to a quantum theory in which probabilities are conserved. The post also leaves open the definition of the interacting theory in continuous spacetime, pointing to OpenAI’s recent work on Yang-Mills theory as a related piece. TSN has already covered OpenAI’s earlier public claims about AI-assisted work on the Exponential Time Hypothesis and Yang-Mills.
A candidate preprint is not a settled definition
The honest reading is narrower than a solved fifty-year problem. Craig and Astra have posted a candidate nonperturbative lattice definition, in a guest blog and in an October preprint, with OpenAI listed as an affiliation. The blog is explicit that Astra led the detailed proof after Craig supplied the strategy. It is also explicit that the work has not closed every question the construction needs to answer. Reflection positivity at finite lattice spacing is still open, by the author’s own account, and so is the interacting theory in the continuum. Those are not minor loose ends if the goal is a definition that can be used in simulations and trusted as a quantum theory.
The paper has not been peer reviewed. The three lattice theorists named in the post looked over the proof and asked questions; that is valuable scrutiny, and it is not the same as acceptance by a journal. Until other specialists test the construction, and until the open points Craig flags are resolved or clearly bounded, the result should be held as a candidate, which is the word both the blog and the arXiv abstract use.
This sits in a wider run of laboratory claims about AI-assisted proofs, including Microsoft’s account of human-directed agents helping to prove the KLS conjecture and OpenAI’s earlier publication of hundreds of AI maths and computer-science results, some of them Lean-checked. Those earlier stories are useful context. They leave this preprint short of a finished theory. The facts on the record are a guest post, a candidate lattice definition, a set of author caveats, and an open question about reflection positivity.
Sources
- Nathaniel Craig, “A step toward defining the Standard Model on a lattice,” OpenAI science blog, 15 November 2026
- Nathaniel Craig, “A gauge-invariant measure for lattice chiral fermions,” arXiv:2610.09093, October 2026
Get TSN’s biggest tech stories, explained.
News and insights across AI, tech, crypto and cybersecurity. Every source linked. Unsubscribe anytime.
Get the newsletterRelated stories
- OpenAI Says Its AI Refuted the Exponential Time Hypothesis is the nearest TSN account of OpenAI’s earlier public claims about AI-assisted work on hard maths and Yang-Mills, and the right next stop for that context.
- Microsoft Says Human-Directed AI Agents Helped Prove the 1995 KLS Conjecture
- OpenAI Publishes 372 AI Maths Results; About 42% Are Lean-Checked
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X