What happened
On September 16, Danish drugmaker Novo Nordisk and Anthropic announced a collaboration in Bagsværd: Novo will use Anthropic's research-oriented Claude Science platform and frontier models to accelerate the discovery and development of new medicines, while also using the models to strengthen AI-driven software development internally.
The companies will jointly address drug-discovery challenges identified by Novo's scientists and computational teams, building targeted solutions and workflows to support what the announcement calls biological reasoning. The first step is a trial run — Claude Science tested in selected R&D workflows, not an enterprise-wide rollout.
Governance in the headline paragraph
Unlike most "AI empowers R&D" announcements, the constraints sit in a prominent place: the work proceeds under data governance and human oversight, following Novo's own ethics and compliance standards.
One concrete number explains why Novo is willing to keep betting: according to Unite.AI, its earlier NovoScribe deployment cut clinical study documentation from more than ten weeks to ten minutes. Documentation is not a drug, but it is one of the most labor-intensive parts of running a clinical trial.
The sober part
Big pharma's use of AI is moving from document processing and patient screening toward drug discovery and scientific research. The finish line has not moved: no AI-discovered drug has been approved to date, and every candidate molecule still has to survive clinical trials and regulatory review.
What is worth tracking here is not the announcement language but what happens after the trial — which specific scientific questions Claude Science stays on, and which ones it gets pulled from. The history of pharma AI procurement is littered with pilots and short on tools that reached the core research workflow. Writing human oversight into the contract suggests Novo knows exactly where that line is.
[1][2]