Research labs are confronting a new compliance hurdle as regulators clarify who must sign off on AI‑generated scientific data. Anthropic’s Claude Science, an AI workbench launched in June 2026, enables multi‑agent workflows that can design CRISPR screens, predict protein structures and perform chemoinformatics without a human directly authoring the final record.
Regulatory signals from Europe and the United States
The European Commission’s 2025 draft GMP Annex 22 draws a firm line around adaptive AI models, barring them from GMP‑critical tasks unless a human can intervene and provide documented oversight. The draft also places responsibility for GMP compliance on the model’s vendor, a provision that could affect Claude Science if it is used in critical laboratory functions.
In the United States, the Food and Drug Administration issued a warning letter in April 2026 to Purolea Cosmetics Lab after it used AI agents to draft drug specifications and master production records without subsequent human review. The FDA cited a violation of 21 CFR 211.22(c) and emphasized that all AI‑generated results must be reviewed and approved by a qualified individual in the quality‑control unit before becoming an official record, as required by Section 501(a)(2)(B) of the Federal Food, Drug, and Cosmetic Act.
What the rules mean for AI‑assisted research
Both regulatory developments focus on the need for a secure, time‑stamped audit trail tied to an authenticated human identity, as mandated by 21 CFR Part 11. While Claude Science provides reproducible code and data provenance, it does not automatically satisfy the human‑signature requirement for official records.
Companies using agentic AI must therefore establish clear delegation chains. When a coordinating agent assigns tasks to sub‑agents, the final output—typically the coordinator’s summary—must be reviewed and signed by a qualified researcher or quality‑control professional. Without that step, the record may not meet FDA or future EU GMP standards.
Industry response
Vendors are already working to embed compliance features into their platforms, such as built‑in human‑approval checkpoints and detailed logs of agent delegation. Anthropic has announced funding for up to 50 AI‑for‑Science projects through December 2026, indicating that many labs will adopt these tools while navigating the evolving regulatory landscape.
As AI becomes a routine part of scientific R&D, the question of accountability remains central. Organizations must align their workflows with both existing electronic‑record regulations and the emerging expectations of regulators on both sides of the Atlantic.
Original reporting: El Paso News (HLL/CB) — read the source article.
