AI Outperforms Virologists, Raising Bioweapon Concerns

Thumbnail Image

The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.

A study by researchers from the Center for AI Safety, MIT Media Lab, UFABC, and SecureBio shows that AI models like ChatGPT and Claude outperform PhD-level virologists in advanced lab troubleshooting tests. The findings raise dual-use risks, suggesting that such technology could be misapplied to develop dangerous bioweapons.[AI generated]

Why's our monitor labelling this an incident or hazard?

The article explicitly involves AI systems (ChatGPT, Claude, OpenAI's o3, Google's Gemini 2.5 Pro) demonstrating advanced capabilities in virology-related problem-solving. While no direct harm has occurred yet, the potential misuse of these AI models by non-experts to create deadly bioweapons represents a credible and plausible risk of harm to health and safety. Therefore, this event fits the definition of an AI Hazard, as it could plausibly lead to an AI Incident involving injury or harm to people.[AI generated]
AI principles
SafetyRobustness & digital securityAccountabilityRespect of human rightsTransparency & explainability
Industries
Healthcare, drugs, and biotechnologyGovernment, security, and defenceDigital security
Affected stakeholders
General public
Harm types
Physical (death)Physical (injury)Public interestHuman or fundamental rights
Business function:
Research and development
AI system task:
Content generationInteraction support/chatbotsReasoning with knowledge structures/planning

Articles about this incident or hazard