The brief · 19 May 2026

The AI brief, 2026-05-19

This Tuesday, Google finally publishes the lab results for its Co-Scientist in Nature. Brussels, meanwhile, is clarifying what it really means by "high-risk AI."

1 min read J / K to navigate

Google Publishes Co-Scientist in Nature, With Lab-Validated Biological Targets

Co-Scientist doesn't just spit out hypotheses. Multiple Gemini-based AI agents debate them internally before handing them off to researchers, who then tested some of the leads at the bench. Across acute myeloid leukemia, liver fibrosis, and antimicrobial resistance, the system's hypotheses held up against real-world experiments.

practitioners › Published May 19, 2026 in Nature, with Google DeepMind: three independent biomedical validations, including a drug repurposing candidate for AML.

www.nature.com →

Brussels finally releases examples to clarify what counts as high-risk AI

For months, companies deploying AI in Europe have had to guess whether their system fell into the AI Act's high-risk category. The Commission has now published draft guidelines with concrete cases for providers, deployers and authorities, which should clarify who needs to comply with the heaviest obligations.

practitioners › The consultation stays open until July 23, 2026: this is the window to flag your sector's borderline cases before the classification gets locked in.

Commission européenne →