Asilomar Principles for Managing AI Applications in Nuclear and Biosecurity

June 18, 2026
Alison Burke, Stephen Herzog, Yanliang Pan, William Potter, Douglas Shaw

Our Secretariat adopts the following principles, based on discussions with more than 100 experts from academia, research institutions, government, civil society, and the artificial intelligence (AI) industry. These principles are intended to guide the responsible governance of AI applications that impact nuclear and biological security. They recognize not only AI’s potential contribution to human security, but also its ability to create or amplify catastrophic risks on a global scale. As the first statement of the ongoing Asilomar process, the principles are intended to set the agenda for further research and implementation work, leaving room for improvement as experience and capabilities evolve.

1. AI must protect human survival.

AI systems must strengthen and never erode barriers to the use of nuclear and biological weapons. These weapons pose an extinction risk to humanity even before the development of AI. Such risks must not be accelerated or exacerbated by AI systems. Therefore, protecting human survival must be a top priority in the deployment of AI tools that impact these areas.

2. The decision to use nuclear weapons must be subject to meaningful human control.

AI systems must not initiate, authorize, or cause the use of nuclear weapons. Human decision makers must maintain the ability to review and override AI output, even under intense time pressure and in situations where automation biases can skew their decisions. Therefore, AI systems involved in nuclear decision support must be auditable by civilian and military authorities, both in data and logic, both in peacetime and in crisis. New intelligence, surveillance, and reconnaissance (ISR) systems and nuclear command, control, and communications (NC3) architectures should be deployed only if proven to reduce the risk of nuclear weapons use.

3. AI governance must enhance non-proliferation and strategic stability.

Given the disruptive potential of AI, nuclear and biological research activities need to become safer, more secure, and more resistant to proliferation. Existing containment practices must evolve to address the new risks posed by AI, including efforts to reduce the risk of AI-induced escalation, miscalculation, and proliferation. Behavioral arms control and confidence-building measures should be pursued alongside the responsible use of AI tools to improve crisis communication.

4. AI developers must contribute to predictive risk governance.

Frontier AI developers have a special responsibility to help predict and manage nuclear and biological risks that can arise from rapid commercial innovation. Advances in AI could bring technological shockwaves to these areas before governments and international organizations are ready to manage them. Therefore, developers should evaluate new features before release, recognizing how they may change incentives or lower practical barriers to weapons use and proliferation, and support stronger oversight in response to increased risk. AI companies are geopolitical actors whose choices can impact global security. Their legal, financial and institutional obligations should reflect the overriding priority of preventing the risk of extinction.

5. AI-powered monitoring and verification must be responsible and ethical.

AI systems have the potential to significantly improve efforts to monitor and verify peaceful nuclear and biological activities and detect diversion in support of weapons of mass destruction programs. Because these decisions carry significant risks, the use of AI must not undermine established standards of explainability, objectivity, validity, data provenance, and ultimate human accountability. AI must be used in a way that protects privacy and personal safety while preventing the disclosure of sensitive nuclear or biological information that could aid malicious actors or undermine strategic stability. AI models used for monitoring and verification must protect themselves from becoming tools for spreaders to evade detection.

6. AI governance must be globally inclusive.

International cooperation, aligned with frameworks such as the Treaty on the Non-Proliferation of Nuclear Weapons (NPT) and the Biological Weapons Convention (BWC), is needed to ensure that the benefits of AI reach all humanity without deepening security and development gaps. This effort must foster a common strategic understanding and reduce, rather than exacerbate, the risks of nuclear proliferation, nuclear escalation, war, and catastrophic biological events. Measures to limit access to dangerous capabilities therefore need to be combined with efforts to reduce incentives for states and other actors to acquire them.

7. AI must not enable disinformation or attacks on nuclear or biological facilities.

False or manipulated information about the use or development of nuclear or biological weapons can cause great damage. Safeguards must be established to prevent attackers from using AI to create or spread very real falsehoods in these areas. We must also develop proactive resilience against AI-enhanced physical and cyber attacks on nuclear and biological facilities, including attacks that enable material theft and sabotage. These measures need to address both crisis decision-making and public perceptions of nuclear and biological threats. Reversing the social, economic, and psychological effects of information warfare may be difficult.