Researchers propose layered safeguards for AI biological design
A team led by University Medical Center Mainz recommends security checks spanning model development, access to sensitive data and biological production.

An international research team led by University Medical Center Mainz has proposed multiple layers of safeguards for generative AI used to design biological sequences. Published in Nature Communications, the recommendations address potential misuse while seeking to preserve medical and biotechnology research.
The researchers stress that current evidence does not show AI can create wholly new human pathogens that are more dangerous. Their concerns focus instead on future advances: more capable models could lower some barriers between a biological idea and a real threat.
The proposed approach would extend protection across training datasets, model development and biological production, rather than relying on one safeguard. For sensitive information about pathogens, the team favors graduated access permissions over a choice between fully open and fully closed access.
Other recommendations include identity checks, closer tracking of critical synthesis equipment and stronger obligations for providers. The researchers also call for model operators to assess risks throughout a model’s life cycle, preferably with independent experts, and for funding bodies to consider biological safety when evaluating projects.
The proposals grew out of a three-day workshop convened by Wübben Stiftung Wissenschaft in November 2025. It brought together 20 experts spanning molecular biology, AI, bioinformatics, ethics, medicine and politics. The authors caution against broad restrictions that could unnecessarily impede responsible research.




