Dario Amodei, chief executive of Anthropic, addressed the United Nations Security Council via video link on September 23, 2026, urging member states to begin negotiations on a binding agreement to prohibit the development and use of artificial intelligence in creating biological weapons. Speaking from Anthropic’s headquarters, Amodei emphasized the urgency of establishing clear international norms before AI capabilities advance to a point where misuse becomes difficult to detect or prevent. His appeal focused on starting with narrow, achievable agreements rather than waiting for comprehensive treaties that may take years to finalize.
Amodei argued that AI’s ability to accelerate pathogen design, optimize transmission mechanisms, or evade existing detection systems presents a novel and growing threat to global security. He noted that while current AI models are not yet capable of autonomously designing viable bioweapons, the trajectory of progress demands preemptive governance. The Anthropic CEO stressed that waiting for proof of harm would be too late, citing the dual-use nature of AI research in life sciences where beneficial applications can be rapidly repurposed. He called on the Security Council to leverage its authority under Chapter VII of the UN Charter to initiate talks on a prohibition framework, similar to existing treaties banning chemical and biological weapons.
Amodei explained that attempting to regulate all AI applications in biology at once would be impractical and could hinder legitimate scientific research. Instead, he proposed focusing initial efforts on a clear red line: the use of AI to design or enhance biological agents intended for hostile purposes. This approach, he said, mirrors past successes in arms control where specific prohibitions paved the way for broader agreements. He acknowledged verification challenges but suggested that monitoring AI training data, model outputs, and laboratory workflows could form part of a compliance system. The goal, Amodei said, is to establish an international stigma against AI-enabled bioweapon development, making it politically and technically costly for states or non-state actors to pursue.
Amodei acknowledged concerns that overly broad restrictions could impede pandemic preparedness or drug discovery efforts. He responded that any ban must include exemptions for defensive research conducted under strict multinational oversight, such as projects authorized by the World Health Organization or conducted in secure, transparent facilities. He pointed to the Biological Weapons Convention’s existing confidence-building measures as a model for transparency without sacrificing security. The Anthropic CEO also proposed creating an international scientific advisory panel to help distinguish between benign and malicious AI use in biology, ensuring that policy keeps pace with technological change. He warned that fragmentation in national approaches would create loopholes that malicious actors could exploit.
What specific actions is Dario Amodei asking the UN Security Council to take? Amodei is urging the Council to initiate formal negotiations on a binding international agreement that prohibits the use of artificial intelligence in the design, development, or enhancement of biological weapons, starting with a narrowly defined ban to build consensus.
How does Amodei suggest verifying compliance with an AI bioweapon ban? He proposes monitoring AI training datasets, model outputs, and laboratory activities, combined with transparency measures and multinational oversight for legitimate research, drawing inspiration from existing arms control verification frameworks.
Why does Amodei believe a narrow ban is more effective than waiting for a comprehensive treaty? He argues that focusing on a clear prohibition first allows for quicker agreement and establishes an international norm, while broader regulations can follow once foundational consensus is reached, avoiding delays that could allow threats to emerge.