Anthropic CEO Proposes Global AI 'Speed Limit' to Curb Catastrophic Risk
Dario Amodei urges leading AI labs to decelerate development, but warns that geopolitical rivalry with China makes global coordination a 'tough dilemma'.
Anthropic CEO Dario Amodei has called on the world's leading AI companies to decelerate the development of their most powerful models to prevent catastrophic risks. The proposal seeks to establish a global "speed limit" on AI progress to ensure safety protocols keep pace with raw capability.
In a recently published essay, Amodei outlined a three-step framework for this slowdown. The plan begins with granting third-party evaluators employee-level access to verify model safety, followed by the coordination of safety standards among democratic nations, and finally, establishing coordination between democratic and authoritarian governments. Anthropic has already unilaterally committed to the first step regarding third-party verification.
The Geopolitical Standoff
Despite industry interest, Amodei identified China as the "toughest dilemma" in the pursuit of a slowdown. He argued that the military and strategic incentives for adversarial nations to ignore a speed limit are immense, stating that the advantage gained from pulling ahead is too large to easily overlook. This tension persists as Chinese startups continue to gain ground against U.S. leaders in a high-stakes competition for AI supremacy.
On the other side of the divide, Chinese President Xi Jinping has signaled an interest in governance, calling for a "consensus-based global AI governance framework" and the creation of a BRICS AI open-source community during a summit in India. These diplomatic signals arrive as the world anticipates upcoming meetings between President Xi and U.S. President Donald Trump.
Industry Friction and Risks
The push for a slowdown faces both internal and external criticism. Jacob Coxon, a researcher at Anthropic, recently resigned from the company, claiming that both OpenAI and Anthropic are "gambling with our lives." From a commercial perspective, David Sacks has challenged the altruism of the proposal, arguing that the drive for a slowdown is actually motivated by "massive product-liability exposure" rather than a pure desire for safety.
Why It Matters
The outcome of this debate will determine if AI development is governed by human values or a geopolitical arms race. If democratic nations unilaterally slow their progress while authoritarian regimes do not, the U.S. risks losing a critical economic and military edge. However, if no global coordination is achieved, the risk of an unaligned superintelligence causing a global catastrophe increases significantly.
What's Next
Observers are now watching to see if the proposed safety standards can be codified into formal agreements between the U.S. and its allies. The effectiveness of the plan remains unconfirmed, as it depends entirely on whether adversarial governments are willing to sacrifice strategic advantages for collective safety. Additionally, the potential IPOs of OpenAI and Anthropic may introduce new commercial pressures that conflict with the goal of a deliberate slowdown.