Anthropic says scientist used Claude Science to build first complete ultraviolet sky map

Anthropic says a scientist used Claude Science, its new AI research workbench, to produce the first complete map of the sky in ultraviolet light.

The method was not a new telescope or a fresh survey. The researcher combined existing ultraviolet datasets and used the tool to fill in the regions those datasets never covered.

What Anthropic says happened

Claude Science launched on June 30, 2026. Anthropic describes it as an AI workbench aimed primarily at computational biologists, with the flexibility to handle other fields such as astronomy.

The product bundles several stages of research into one place. It covers literature review, data analysis, experiment planning, figure generation and scientific writing.

All of that runs inside a reproducible environment. In practice, another researcher should be able to retrace the steps that produced a result, which matters a great deal in science.

According to Anthropic, the ultraviolet project used that toolkit to stitch together prior observations into one map. The software then handled gap-filling for the parts of the sky that earlier missions had left blank.

Anthropic has not publicly named the scientist behind the work. It also has not listed the specific datasets that went into the final map.

A long-running gap in the sky

The best-known effort came from NASA’s Galaxy Evolution Explorer, better known as GALEX. Its data, published in 2010, covered approximately 75% of the sky.

GALEX observed in two ultraviolet bands. Its far-ultraviolet band sat at 1530 Å, and its near-ultraviolet band sat at 2310 Å.

In 2021, a separate project used deep learning to build combined all-sky maps from FIMS data, an effort that underlined how patchy ultraviolet records had been.

In March 2026, researchers announced a 3D map of Lyman-alpha ultraviolet emission from the early universe, a project unrelated to Claude Science.

What this means

For astronomers, a complete ultraviolet map would be a useful reference. Researchers studying star formation, galaxies and other ultraviolet sources would have one continuous view instead of a patchwork with holes.

Observed data and inferred data are not the same thing. A credible all-sky product would need to show which pixels came from an instrument and which came from a model.

Until the researcher and the source datasets are identified, that question stays open.

The reproducible environment that Anthropic built into Claude Science could help here. If the workflow is shared, other scientists could rerun it, inspect the assumptions and test the gap-filled regions against new observations.

Three things are worth watching from here. First, whether Anthropic or the researcher publishes the underlying datasets and methods. Second, whether independent astronomers compare the gap-filled areas with other instruments. Third, whether similar claims start appearing from biologists, the audience Claude Science was actually built for.