Biosecurity

Definition

Biosecurity encompasses the strategies, regulations, and technical safeguards intended to prevent the development, acquisition, or use of biological agents for harmful purposes. In the context of artificial intelligence, biosecurity concerns have expanded to address the risk that AI systems, particularly large language models and protein structure prediction tools, could lower the barrier to designing or synthesising dangerous pathogens. AI capabilities in molecular biology, including protein folding prediction, gene synthesis optimisation, and literature synthesis, create dual-use risks where tools developed for beneficial research could be repurposed by malicious actors. Biosecurity governance must therefore account for both traditional biological threats and the novel acceleration pathways that AI introduces.

How It Relates to AI Threats

Biosecurity intersects directly with the Systemic-Catastrophic threat domain through the ai-assisted biological threat design sub-category. AI systems capable of predicting protein structures, suggesting genetic modifications, or synthesising dispersed scientific knowledge could enable actors with limited expertise to design dangerous biological agents. This represents a paradigm shift in biosecurity threat modelling, as traditional barriers of tacit knowledge and specialised laboratory access may be partially circumvented through AI assistance. The concern is not hypothetical; evaluations of large language models have demonstrated the ability to provide step-by-step guidance that could assist in biological weapon development.

Why It Occurs

• AI tools for molecular biology and genomics are increasingly powerful and publicly accessible
• Large language models can synthesise dispersed technical knowledge into actionable biological instructions
• Traditional biosecurity frameworks were designed before AI lowered knowledge barriers to pathogen design
• Protein folding prediction and gene synthesis tools have legitimate scientific uses that resist restriction
• Verification and screening mechanisms for DNA synthesis orders have not kept pace with AI capabilities

Real-World Context

No incidents in the TopAIThreats database currently document AI-assisted biological threats, though the risk is recognised as a priority by major governments and AI laboratories. In 2023, researchers demonstrated that large language models could provide information potentially useful for biological weapon development, prompting enhanced safety evaluations. The U.S. Executive Order on AI Safety (October 2023) specifically addresses dual-use biological risks. AI laboratories including Anthropic and OpenAI have implemented biosecurity-specific red-teaming and content restrictions. The Nuclear Threat Initiative and Johns Hopkins Center for Health Security have published frameworks for AI-biosecurity governance.