A coalition of more than 100 researchers from leading institutions including Johns Hopkins, Oxford, Stanford, Columbia and NYU is urging policymakers to tighten safeguards on certain high-risk biological datasets to prevent misuse by artificial intelligence. Their concern stems from the possibility that advanced AI models could be used to analyse and exploit openly available infectious-disease data to design or enhance dangerous viruses, presenting a significant gap in current biosecurity frameworks.
The experts argue that once sensitive pathogen data is publicly accessible on the internet, it becomes nearly impossible to retract or fully control, meaning that third parties could adapt powerful AI systems to pursue risky biological engineering. While mainstream commercial models like ChatGPT or Claude are not yet able to create actual biological threats on their own, the researchers warn that combining freely available data with capable AI tools could lower barriers to misuse if proper protections are not introduced.
To address this challenge, the group proposes a governance framework similar to how sensitive health records are protected, balancing scientific openness with necessary restrictions. This could involve categorising and restricting access to datasets that pose higher biosecurity risk while still enabling legitimate scientific research — an approach that does not currently exist in most open data environments.
The call for action emphasises that there is a “window of opportunity” to enact safeguards before AI capabilities evolve further, outpacing current safety measures. Without timely policy development and data governance frameworks, the risk of AI-assisted biological misuse — including the potential creation of harmful agents — could grow rapidly, prompting experts to stress proactive regulation rather than reactive responses.
