Anthropic CEO urges slower frontier AI push

On 12 September 2026, Anthropic CEO Dario Amodei released an essay urging an immediate slowdown in the pace at which frontier artificial intelligence model capabilities are improved. He said the issue is not whether to develop AI, but that recent advances have begun to outstrip the ability of companies and governments to understand and control the technology. Amodei argued that AI’s growing role in helping to design and train newer, more capable systems has accelerated since around the summer, raising the risk that human oversight could fall behind.

Amodei framed his proposal as a way to build AI at a balanced rate that preserves potential benefits while giving safety work time to catch up. He called the risks associated with frontier systems "serious" and said that slowing the pace of capability gains would buy researchers and policymakers more time to develop safeguards.

The essay outlines a three-point plan intended to govern the next phase of frontier AI development. First, Amodei proposed independent third-party monitoring of AI models as they are developed, with evaluators embedded closely enough to observe systems in real time. This monitoring would track capabilities and potential risks as models scale, rather than relying only on occasional external audits.

Second, he called for industry-wide standards or regulation to guide how powerful models are trained, evaluated and deployed. Amodei argued that shared rules would help ensure that competitive pressures do not drive companies to cut corners on safety. Third, he urged international coordination, including measures involving authoritarian governments, to reduce the chance that divergent national approaches undermine global risk management.

To demonstrate that the proposed monitoring can be put into practice, Amodei said Anthropic will unilaterally give outside evaluators "ongoing, employee-like access" to its operations. This access is set to include desks, access badges and company laptops so that evaluators can work alongside internal teams.

Amodei also urged other frontier AI companies to match this level of transparency and access. By embedding independent evaluators within development teams across the industry, he argued that potential harms and failure modes could be identified and addressed earlier in the development process.

As the proposal circulated on 12 September, leaders at other prominent AI companies publicly voiced support. OpenAI CEO Sam Altman wrote that he agreed with Amodei about the need to "pace the frontier" of AI development. Elon Musk also posted that "Dario is right," aligning himself with the call for a more deliberate approach to advancing cutting-edge systems.

Their comments indicate a degree of convergence among key figures in the sector around the idea that frontier AI progress should be slowed to allow safety measures to catch up. Amodei’s plan combines this call for pacing with specific governance mechanisms, including embedded monitoring, regulatory standards and international cooperation.