Prioritizing Feasible and Impactful Actions to Enable Secure AI Development and Use in Biology.

Area of Science:

• Biotechnology and Artificial Intelligence
• Biosecurity and Risk Management

Background:

• Artificial intelligence (AI) offers significant potential for biological innovation.
• However, the risk of AI-enabled biothreat creation necessitates proactive mitigation strategies.
• Current actionable steps to address these risks are underdeveloped.

Purpose of the Study:

• To identify and prioritize actionable policy and technology strategies for mitigating AI-enabled biothreat creation risks.
• To provide a framework for understanding and addressing risks across the AI-enabled biothreat lifecycle.
• To foster collaboration among stakeholders for developing layered defense mechanisms.

Main Methods:

• Collated existing reports and mitigation options from diverse sources.
• Evaluated proposed mitigation options based on projected feasibility and impact.
• Organized strategies within a framework aligned with AI-enabled biothreat creation steps.

Main Results:

• Prioritized seven key mitigation strategies, encompassing twelve individual mitigations.
• Key strategies include model unlearning, input/output filtering for biological design tools (BDTs), AI agents for biosecurity, safety bug bounties, enforcement of material protections, enhanced biosurveillance, and metadata screening before DNA synthesis.

Conclusions:

• A layered defense approach is crucial for ensuring the safe and secure use of AI in biotechnology.
• Collaboration between policymakers, researchers, and technologists is essential for refining and implementing these mitigation strategies.
• Proactive development and implementation of these strategies will help unlock the societal benefits of AI in biology while minimizing misuse potential.