On October 8, 2026, Anthropic published “The missing map of the sky”, in which Brice Menard, an astrophysicist at Johns Hopkins University and a researcher at Anthropic, describes using Claude Science to produce what he calls the first complete map of the sky in ultraviolet light (far-UV 154 nm and near-UV 232 nm combined).
UV light is blocked by the atmosphere, so it can only be observed from space, and existing maps are full of holes. NASA’s GALEX mission imaged about two-thirds of the sky in some 38,000 observations between 2003 and 2013 but skipped bright regions, including much of the Milky Way’s plane, to protect its detectors. Claude orchestrated a team of agents that found and downloaded the public UV surveys, removed glare around bright stars, cross-calibrated the instruments onto one scale and grid, and filled the roughly one-third of the sky never observed in UV. That gap-filling learned how UV brightness relates to visible, infrared and radio maps and predicted the missing regions with per-pixel uncertainties. In hold-out tests the predictions matched hidden real UV data to within about 10 percent. Light from more than 100 million stars measured by ESA’s Gaia was added on top.
The process was not automatic. Menard spotted faint circular artifacts, the footprints of individual GALEX observations left by residual atmospheric glow, which had passed two rounds of review by other agents. Claude then corrected all 38,000 observations. The map went through more than a dozen versions over several days, with Claude running hours of computation between exchanges.
Why it matters: this is the kind of useful but low-priority project that scientists rarely find weeks for. It shows agents doing the long, careful calibration work under expert steering, a different contribution from proving a theorem.
What it does not show: about a third of the map is a statistical prediction, not a measurement, and the map labels which pixels are which. It is a first-person account from an Anthropic researcher, not a peer-reviewed paper, and it took human checking to catch a systematic error the agents missed.