Mathematical Community Calls for Responsible Release of AI-Generated Mathematics

The recommendations address the risks posed by proprietary models and the potential for a "two-tier system" where frontier research is inaccessible to the broader scientific community. The community emphasizes that current practices, where AI outputs mathematical arguments that users may not fully understand or be able to verify, challenge established scholarly norms of peer review and human understanding.

The proposed guidelines offer two distinct paths depending on the level of human involvement:

• For results accompanied by a responsible mathematician who understands the content, traditional academic norms such as preprints, peer review, and seminars should be followed.
• For AI-generated outputs not understood by the prompter, the group recommends several technical and institutional norms.

Technical recommendations include citing relevant literature, presenting proofs in standard mathematical conventions, and depositing results in appropriate, non-proprietary scholarly repositories. The guidelines also suggest that AI labs should disclose the model name, prompts used, chain of thought, computation time, and estimated costs for each released result. Furthermore, they encourage the formalization of proofs to meet community standards.