Oct 28, 2026

OpenAI Claims AI Proofs in Math and Physics; Oracle and Google Push AI Further

OpenAI says its models proved major results in math, computer science and physics, now under expert review. Oracle embeds ChatGPT and Codex in its apps, and DeepMind unveils Genie 4.

OpenAI is asking the scientific world to check what could be the most consequential set of claims an AI lab has ever made: that its models proved a decades-old conjecture in harmonic analysis, refuted one of theoretical computer science's foundational assumptions, and produced a rigorous existence proof for the quantum field theory behind the strong nuclear force. Elsewhere, Oracle and OpenAI deepened their alliance at Oracle AI World, and Google DeepMind unveiled Genie 4, its most capable interactive world model yet.

OpenAI puts three AI-made results up for review

OpenAI announced this week that its models generated results in three fields, and it is handing each one to outside experts rather than declaring victory on its own. The company has not said which models produced the work.

Mathematics: local smoothing in every dimension

The first claim concerns the local smoothing conjecture, posed by Christopher Sogge in 1991, which describes how concentrated waves spread out over time. Mathematicians had settled it only in low dimensions; OpenAI says its models proved it in every spatial dimension, and that the result also resolves the Kakeya and Stein restriction conjectures. According to Interesting Engineering, the proof was generated on October 18 and formalized in the Lean proof assistant the following day, with both versions posted to GitHub. Alex Iosevich of the University of Rochester, whose doctoral adviser formulated the conjecture, told Al Jazeera that OpenAI appeared to have cracked it, calling the moment historic.

Computer science: the Exponential Time Hypothesis

The second claim is that the models disproved the Exponential Time Hypothesis (ETH), the widely assumed idea that certain satisfiability problems cannot be solved meaningfully faster than brute force. ETH underpins conditional hardness results in hundreds of papers across graph theory, databases, computational biology and geometry, so a valid refutation would force a broad reassessment. Russell Impagliazzo and Ramamohan Paturi of UC San Diego, who introduced the hypothesis, will lead the review alongside MIT's Ryan Williams. Williams told Al Jazeera his team is still checking the proof and that practical applications are unlikely in the short term, even though the result would be a major development if it holds. OpenAI has also agreed to fund a conference on the claim.

Physics: quantum chromodynamics and Yang-Mills

The third and boldest claim is a mathematically rigorous existence proof for interacting four-dimensional gauge theories with fermions, including quantum chromodynamics. The work also touches pure Yang-Mills theory, part of one of the Clay Mathematics Institute's Millennium Prize problems. A group led by Nobel laureate David Gross, with Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov, will write its own account before the model-generated physics materials are released, so this result cannot yet be evaluated publicly.

Why the rollout matters

OpenAI calls its expert partners "shepherds" and says the three collaborations operate independently, with participants free to publish what they find. The company acknowledged that choosing the first reviewers creates an information asymmetry and invited feedback on the process. It also pointed to a previously announced $5 million contribution to the Simons Foundation to fund workshops examining the work. Until those reviews conclude, all three results remain OpenAI's claims rather than accepted science, and the ETH and physics results in particular are unverified.

Oracle wires ChatGPT and Codex into its business software

At Oracle AI World in Las Vegas, Oracle co-CEO Clay Magouyrk shared the stage with OpenAI CEO Sam Altman to announce a set of integrations. A new Oracle AI Workspace uses plugins to connect ChatGPT to Fusion Applications, NetSuite, Oracle AI Database and Oracle Cloud Infrastructure (OCI), letting employees query business data and trigger approved workflows from a chat window. Oracle App Composer lets users build enterprise apps with Codex that Oracle then deploys and runs on OCI, and an Oracle AI Security Mesh service will use OpenAI's cyber models to find and fix vulnerabilities. A small-business bundle combines ChatGPT Business, Codex, OCI and NetSuite. None of the products has a firm general-availability date.

Oracle also broadened the hardware on its cloud. Beyond Nvidia Vera Rubin NVL72 systems, which it plans to scale to multi-gigawatt capacity, OCI will offer Google TPUs and Cerebras inference starting in 2027, and Oracle plans to deploy OpenAI's Jalapeño inference accelerator, built with Broadcom. Oracle describes itself as the first hyperscaler outside Google to offer TPUs.

Google DeepMind unveils Genie 4

Google DeepMind introduced Genie 4, a real-time interactive world model. Users move through generated scenes with keyboard controls and reshape them with text prompts, such as changing the weather or adding objects and people. The model simulates physics like collisions and aerodynamics, animates characters that react to their surroundings, and generates ambient audio and live dialogue alongside the visuals. Gemini models help keep scenes coherent and, in robotics demos, break tasks into steps that guide the simulation.

Genie 4 is a research preview for trusted testers for now, with access coming to Google AI Ultra subscribers through Project Genie and to early partners in the Gemini API and AI Studio. DeepMind says all output carries SynthID watermarks, and it acknowledges that visual details and object permanence can drift during long sessions.

The takeaway

The day's news captures AI's two tracks: a push into fundamental science that now hinges on human verification, and a commercial race to embed models and chips into the infrastructure businesses already use. The coming weeks of expert review of OpenAI's proofs will be the story to watch.