Updated

On 13 September 2026, the news section of Science, published by the American Association for the Advancement of Science, reported that a new assessment of the biosecurity risks posed by advanced artificial intelligence had sharply divided the expert community. According to the coverage, some specialists described the report's findings as a "chilling" warning about how AI models could lower barriers to designing dangerous pathogens or toxins, while others argued the conclusions were an overreaction that rested on contested assumptions and limited evidence. The dispute involved biosecurity analysts, AI researchers and policy specialists who disagreed about how much real-world uplift current models actually provide to would-be bad actors. The disagreement matters because such assessments feed directly into how governments and AI developers set safeguards, including model evaluations, access restrictions and rules on publishing sensitive biological methods. Overstating the threat risks constraining legitimate biological research and open scientific exchange, while understating it could leave gaps in oversight as capabilities advance. The report appeared alongside other AI-policy items in the same period, including a US Department of Energy push toward "open" AI models for science, underscoring the tension between openness and security. No consensus emerged from the exchange reported that day.