AI is rapidly transforming life sciences — and not always for the better. Anthropic recently disclosed it disrupted five real-world cases between December and August where actors attempted to use its Claude model to support biological weapons development, including two involving gain-of-function research on dangerous viruses. The company noted that sophisticated state actors routinely try to evade its access controls, and that determining intent is difficult when dual-use research can serve both beneficial and harmful purposes.

The broader picture is alarming. A Stanford team used generative AI to design a synthetic virus — the first time the technology created an organism not found in nature. Researchers from top universities warn that current AI models can already design viral shells, forecast pathogen evolution, and generate DNA sequences that evade biosafety screening. AI leaders including Sam Altman and Dario Amodei have publicly warned of AI-enabled bioweapons, while a survey of 272 AI researchers estimated a 12% chance of AI-enabled catastrophic harm by 2030 — even with mitigation efforts in place.

By the Numbers:

• 5 — cases disrupted by Anthropic involving potential bioweapon-related AI use
• 70% — of 100+ national security experts who believe AI meaningfully raises bioweapon risk now or within 2–3 years
• 12% — estimated probability of AI-enabled catastrophic harm by 2030, even with mitigation efforts
• 20%+ — estimated probability without mitigation efforts

Why it matters: The convergence of increasingly capable AI, accessible gene-editing tools, and a fragmented regulatory landscape is lowering the barrier for bad actors to engineer dangerous pathogens. Experts and policymakers are calling for mandatory government review of frontier AI models for national security risks — before the technology outpaces any hope of oversight.