AI Cannot Be an Author. But Can the Human Claim Its Discovery?

A new tension is emerging between publication ethics and AI-enabled scientific discovery.

JAMA’s current guidance keeps responsibility firmly with human authors. AI may assist with content creation, revision and formatting when its use is disclosed, but humans must verify the content and remain fully accountable. AI may not draft Opinion manuscripts or correspondence, and reviewers must not submit confidential manuscripts to chatbots or similar tools.

Then comes Astra.

OpenAI reports that an internal version of its forthcoming model generated ten results resolving or substantially advancing long-standing problems in mathematics and theoretical computer science. According to OpenAI, the solution-search tokens would have cost roughly $2,000 at Sol API rates. Humans then prepared the manuscripts with the model, which subsequently formalised the arguments in Lean.

OpenAI argues that assigning exclusive human authorship to a proof generated entirely by AI would misrepresent how the discovery was produced.

The apparent conflict may arise because authorship currently combines several distinct functions: discovery, verification, accountability and credit. JAMA answers who must verify the work and accept responsibility. OpenAI asks whether that person or institution should also receive exclusive credit for an intellectual contribution generated by AI.

My takeaway: responsibility must remain traceable to humans and institutions. But scientific attribution should also describe who—or what—produced the decisive insight. As AI moves from assisting communication to generating original contributions, treating discovery and accountability as the same thing may become increasingly difficult.

OpenAI’s report comes from the model developer. The mathematical community must still assess the originality, significance and broader implications of the results.