Chi-kwan Chan, Associate Astronomer at the University of Arizona’s Steward Observatory and Secretary of the Event Horizon Telescope Science Council, used OpenAI’s Codex to refine algorithms for simulating plasma and particle behavior around black holes. OpenAI published the case study on June 10.
The code in question is general-relativistic plasma simulation: coordinate-agnostic, GPU-portable particle-in-cell work that models collisionless plasma in curved spacetime. It is the kind of computation that underpins EHT analysis and empirical tests of Einstein’s field equations. This is not boilerplate.
The same coding-agent capability that accelerated exploit development in the n-day security study is here compressing the algorithm-refinement loop for frontier astrophysics. The capability is neutral. Chan’s work is the clearest recent evidence that Codex is useful at the edge of scientific computing, not just for web-app scaffolding.
Scientists on GPU-heavy simulation codebases should treat this as a signal to run Codex against their own algorithm-testing loops before dismissing coding agents as a developer-tooling story.
OpenAI (openai.com), published 2026-06-10.
