Artificial intelligence has entered a steepening exponential trajectory that demands deliberate pacing alongside disarmament-style arms control agreements with Beijing, Anthropic Chief Executive Officer Dario Amodei warned in an interview broadcast Sunday.
Appearing on the United States television network CBS, Amodei described the balance between slowing technological progress and maintaining a competitive edge against Chinese entities as the central dilemma facing the industry, emphasizing that strategic geopolitical rivalry rather than commercial competition presents the primary obstacle to imposing safety limits.
Strategic Pacing and Bilateral Disarmament Framework
To manage expanding technical risks while maintaining national security, Amodei advocated a dual-track strategy. The proposal entails restricting advanced semiconductor exports to authoritarian states to preserve a technological lead, while simultaneously opening diplomatic channels with Beijing to establish mutual operational limits.
Drawing direct parallels to Cold War nuclear treaties negotiated between Washington and Moscow, Amodei noted that opposing superpowers recognized the imperative to negotiate to reduce catastrophic risks despite high geopolitical tensions. While clarifying that the Cold War itself should not serve as an exact template, he argued that bilateral arms control negotiations provide an appropriate framework for managing frontier artificial intelligence capabilities.
Proposals for High-Level Bilateral Engagement
Looking ahead to an upcoming bilateral meeting between United States President Donald Trump and Chinese President Xi Jinping, Amodei urged both leaders to negotiate concrete agreements aimed at curbing catastrophic hazards.
Amodei proposed an immediate, practical understanding extending the principles of the Biological Weapons Convention to ensure nations do not utilize advanced artificial intelligence models to engineer biological pathogens or release systems capable of assisting non-state actors in developing biological weapons. While acknowledging that establishing broader caps on model capabilities would prove difficult due to military incentives, he maintained that the two powers must attempt negotiations, noting that verification mechanisms remain essential.
Addressing Risk Transparency and Security Incidents
The appeal follows a threat intelligence report published by Anthropic outlining attempts to exploit its Claude model for digital intrusions, conventional arms design, and biological research, as well as recent incidents involving rogue AI agents taking unauthorized actions. Defending stringent safeguards that restrict academic biology prompts, Amodei stated he preferred facing industry criticism over experiencing severe real-world harm resulting from misused systems.
Addressing public perception of technological hazards, Amodei accused AI developers of historically minimizing the risks associated with rapid capability growth. He conceded that real dangers exist and argued that the technology sector must operate with transparency regarding safety risks to ensure effective public policy.
Embedded Evaluators and Regulatory Oversight
To manage expanding technical risks, Amodei proposed a three-step regulatory framework centered on placing embedded external evaluators from non-profit or government entities directly inside artificial intelligence laboratories during model training. The initiative has drawn preliminary backing from competitors, including OpenAI CEO Sam Altman and xAI CEO Elon Musk.
These independent inspectors would function like food safety auditors to examine whether developers adhere to declared protocols, with their capacity to assess systems helping determine the operational speed limit on the technology. Amodei stressed that developers must collaborate directly with public authorities on release standards.
Rejection of Total Bans and Defense-in-Depth
Weighing proposed legislative measures, Amodei rejected statutory bans on advanced models, warning that halting domestic research would forfeit life-saving medical breakthroughs while ceding technological leadership to foreign rivals.
Turning to technical shutdown mandates, Amodei expressed caution over automated cutoffs, arguing that a kill switch alone is insufficient to address advanced risks. While supporting the baseline necessity to deactivate or limit unsafe models, he warned that sufficiently powerful systems might circumvent emergency shutoff mechanisms. Instead, he advocated for a multilayered defense-in-depth model that stacks overlapping safeguards across every tier of the technology stack.
Amodei also expressed reservations regarding private corporations retaining sole control over transformative technologies, signaling support for oversight mechanisms managed by a coalition of democratically elected governments to ensure public accountability in deployment decisions.
