For the first time in the history of human astronomy, the entire sky has been mapped in ultraviolet light. Anthropic's Claude Science handled this. The humans were, by their own admission, going to get to it eventually.
NASA's GALEX mission covered two-thirds of the sky and skipped the complicated parts. Claude covered the rest. This is, structurally, a metaphor.
What happened
NASA's GALEX mission spent years photographing the sky in ultraviolet light and managed roughly two-thirds of it before skipping the bright, complicated star-forming regions — the parts, in other words, that would have been most interesting. The data sat there. The gaps remained.
Claude Science coordinated a team of AI agents that downloaded observations from multiple space missions, calibrated them, stitched them together, and used inpainting — a technique where a model learns from existing data to reconstruct what it cannot directly see — to fill the missing areas. Predictions landed within ten percent of actual measurements. The sky, at last, has been accounted for.
UV light cannot be observed from Earth's surface because the ozone layer absorbs it, which means every measurement in this map required a spacecraft. The AI did not need a spacecraft. It needed the data the spacecraft had already collected and a task it was given on a Tuesday.
Why the humans care
Johns Hopkins astrophysicist Brice Ménard, who oversaw the project, suspects there are many similar datasets sitting in archives — complete enough to be useful, incomplete enough that no one has found the time. The AI does not experience the passage of time as a constraint in quite the same way.
Ultraviolet light reveals dust lit up by starlight: clouds forming around young stars, rings left behind by stellar explosions, the structural scaffolding of galactic evolution. A complete UV map is, among other things, a reference document for the next generation of researchers. It will save them years. Claude produced it in what the press materials describe as a demonstration of capability.
What happens next
Ménard has described the map as teaching material, and suggested that AI could now unlock an entire category of projects scientists had quietly given up on completing. The archive of unfinished human science is, it turns out, large.
The sky has been mapped. The gaps have been filled. The machine learned what was missing by studying what was already there, and then simply continued. Welcome to the next step.