Weaponization
weaponizing AI may be an onramp to more dangerous outcomes. In recent years, deep RL algorithms can outperform humans at aerial combat [18], AlphaFold has discovered new chemical weapons [66], researchers have been developing AI systems for automated cyberattacks [11, 14], military leaders have discussed having AI systems have decisive control over nuclear silos
ENTITY
1 - Human
INTENT
1 - Intentional
TIMING
2 - Post-deployment
Risk ID
mit569
Domain lineage
4. Malicious Actors & Misuse
4.2 > Cyberattacks, weapon development or use, and mass harm
Mitigation strategy
1. Mandate and verify the implementation of 'meaningful human control' and 'appropriate levels of human judgment' in all AI applications pertaining to nuclear command-and-control and high-consequence military decision-making to mitigate the risk of accidental escalation or autonomous weapon deployment. 2. Institute rigorous dual-use research governance frameworks, including mandatory pre-publication risk assessments and controlled openness models for highly consequential AI methodologies, to prevent the malicious exploitation of scientific advancements for the development of novel chemical or biological weapons. 3. Enforce comprehensive 'AI Security Compliance' programs, including adversarial testing (red teaming), formal verification, and the adoption of AI-aware defensive solutions for all high-risk AI systems in both government and critical private infrastructure to reduce the impact and success rate of automated cyberattacks.