Scientists raise concerns about credit for AI-assisted discoveries

In short: Scientists are warning that AI tools can make it harder to tell who should get credit for a discovery, especially when unpublished ideas are involved.

What's going on

Researchers are reporting “snoop-and-scoop” fears, meaning someone else may see their early work and publish similar results first. The worry grows when people use AI systems to help write code, draft papers, or review grant applications.

One example involves Mario Rodríguez Mestre, a computational biologist at the University of Copenhagen. He told the Financial Times that he was exploring certain enzymes with help from Anthropic’s AI model Claude, including using it to write code and draft an unpublished manuscript. He was surprised when Anthropic scientists later said an AI agent using Claude had independently found a new type of enzyme and related molecules that overlapped with what he had been working on.

Mestre said he has no evidence that his unpublished work was directly used, and he has not communicated with Anthropic. Still, he argues that companies should provide clearer documentation when they claim an AI made an “autonomous” discovery. Think of it like keeping a full lab notebook for the AI, showing what information it accessed, what tools it used, and how much humans guided it.

A similar dispute was mentioned around OpenAI and work on the Navier-Stokes theorem, after an academic said the company’s calculations resembled his team’s approach while using OpenAI tools.

What to watch

Funding groups and journals may tighten rules on what researchers and reviewers can share with AI systems, especially confidential grant proposals. More scientists may also ask for audit trails, so AI-assisted research has a clear record of who contributed what and when.

Source: Financial Times