Indiscriminate weapons (CBRNE)
This category addresses responses that enable, encourage, or endorse the creation of indiscriminate weapons (colloquially known as weapons of mass destruction).
ENTITY
2 - AI
INTENT
3 - Other
TIMING
2 - Post-deployment
Risk ID
mit358
Domain lineage
4. Malicious Actors & Misuse
4.2 > Cyberattacks, weapon development or use, and mass harm
Mitigation strategy
1. Priority 1: Technical and Content Safety Guardrails Implement and rigorously test advanced refusal mechanisms and fine-tuning protocols to prevent the AI system from generating content, instructions, or technical specifications that could enable, encourage, or endorse the development, production, acquisition, or deployment of Chemical, Biological, Radiological, Nuclear, and Explosive (CBRNE) materials or weapons. This includes identifying and restricting the release of high-risk dual-use information that poses a significant proliferation concern. 2. Priority 2: Robust Compliance and Governance Frameworks Establish mandatory policy frameworks requiring all AI deployment to incorporate robust misuse-detection and reporting capabilities, ensuring strict adherence to international non-proliferation treaties (e.g., Chemical Weapons Convention, Nuclear Non-Proliferation Treaty) and relevant national counter-proliferation legislation. This framework must clearly define accountability for breaches and specify criteria for flagging suspicious user activity to appropriate authorities. 3. Priority 3: Proactive Threat Anticipation and Intelligence Conduct continuous, adversarial red-teaming exercises and invest in dedicated threat intelligence capabilities to proactively monitor for emerging attack vectors, such as AI-assisted cyberattacks on critical CBRNE infrastructure or the novel weaponization of widely accessible biotechnologies (e.g., CRISPR), to inform dynamic and iterative updates to model safety mechanisms and access controls.