Claude Science AI agents complete first ultraviolet sky map

By Splitfeed EditorialLast updated Oct 9, 2026, 09:30 UTC

Astrophysicist Brice Ménard of Johns Hopkins University led a project, published around October 8 to 9, 2026, that produced the first complete map of the entire sky in ultraviolet light. The work relied on Anthropic's Claude Science, a tool built specifically for scientific research. AI agents autonomously handled a multi-step pipeline: downloading data from several space missions, calibrating that data, and using inpainting to fill gaps where observations were missing. The resulting model predictions averaged roughly ten percent deviation from actual measurements, which Ménard and his collaborators treated as a workable level of accuracy for a map of this scope.

The significance of the result sits at two levels. Scientifically, ultraviolet observations are difficult to collect comprehensively because Earth's atmosphere blocks UV light, meaning the data must come from space-based instruments operated across different missions and eras, each with its own calibration quirks. Assembling those datasets into a single coherent all-sky map had not been done before. The broader question the project raises is whether AI agents can now handle the kind of long, multi-step data pipeline that has traditionally required sustained human expertise, and whether a ten percent deviation is acceptable or limiting depends on the specific science being done with the map. Ménard described the project as research that would not have been completed without AI assistance, a framing that puts the tool's role at the center of the scientific credit question.

None of the three camps, Pro-AI, Anti-AI, or Middle Ground, had published reactions to this story at the time of writing. Pro-AI voices would typically treat this as confirmation that AI agents are ready to accelerate frontier science in ways human researchers working alone could not match. Anti-AI voices would likely press on the ten percent deviation figure and ask whether autonomous pipelines introduce systematic errors that human oversight would catch, particularly for data that will feed downstream research. Middle Ground commentators would probably argue that the result is promising but that Ménard's hands-on leadership illustrates the continued importance of domain experts guiding and validating what the agents produce.

The detail most worth watching is how other researchers use the ultraviolet map in their own work, and whether the ten percent deviation proves acceptable across different scientific applications or prompts calls for a revised pipeline with tighter calibration.