Anthropic chief executive Dario Amodei said the artificial-intelligence industry should slow capability development to give safety measures more time to catch up. Amodei published a three-part framework for stronger checks on advanced AI development. He said Anthropic could implement one part on its own while other elements require broader coordination. Frontier laboratories compete for capital, computing resources, researchers and customers.
The CEO warned that systems capable of coordinating large numbers of agents might emerge within six to 12 months. He argued that an additional year or two could be used to advance alignment research. Voluntary deployment gates can be changed by the companies that create them unless contracts or laws make them binding. International coordination is difficult because governments have different security and economic priorities.
OpenAI’s chief executive told Fortune that his company would delay an initial public offering until next year while focusing on safety. AI safety employees have continued to resign publicly from major laboratories. Agent systems combine a model with tools, memory and repeated actions, creating risks that are not captured by a single prompt response.
A former Anthropic employee described a competitive race in which individual companies fear that slowing alone would cede ground. Earlier calls for industry pauses did not produce a binding global limit. Forecasts about future capabilities remain uncertain even when made by industry leaders. Amodei’s timeline and risk assessment are forecasts, not independently verified predictions of a specific event.
The proposal separates actions a single laboratory can take from measures that require collective enforcement. Internal evaluation thresholds, deployment gates and access controls can be adopted by one company. Limits on the overall pace of frontier development are harder because competitors operate under different laws and incentives. The distinction determines which promises can be checked immediately and which remain contingent on negotiations.
What to watch: The detailed commitments Anthropic adopts under its own framework. Whether rival laboratories or governments agree to coordinated development limits.