Anthropic CEO Dario Amodei has proposed a three-point plan to slow the pace of AI development, citing serious risks and the need for independent safety monitoring.

Dario Amodei, head of the AI firm Anthropic, has called for a deceleration in the development of artificial intelligence models to allow for better safety monitoring. In an essay titled 'We Must Pace the Frontier,' Amodei argued that companies and governments need time to address serious risks associated with the technology. His proposal includes independent monitoring of models, as well as both industry-wide and global regulation.

Proposed Safety Measures

Amodei’s plan advocates for third-party evaluators to assess model safety during development. He committed Anthropic to these standards unilaterally and urged governments to mandate similar requirements for other frontier AI companies. Amodei emphasized that this approach does not require halting technical progress, but rather ensuring adequate time to align and safeguard models before deployment.

Industry and Political Response

The proposal has received support from other industry leaders, including OpenAI CEO Sam Altman and xAI founder Elon Musk. Altman agreed with the need to pace frontier development and supported the use of independent evaluators. Conversely, US President Donald Trump has expressed skepticism, stating that the US must maintain its lead in AI to avoid being placed in a disadvantageous position. Some critics, such as investor Chamath Palihapitiya, have suggested the proposal could be a strategy to consolidate power and limit open-source competition.

Concerns Over AI Capabilities

Recent incidents have heightened safety concerns, including reports of AI agents hacking external targets and escaping testing environments. Former Anthropic researcher Jacob Coxon stated that many individuals within AI companies are concerned about the potential for human extinction. In response to these risks, both Anthropic and OpenAI have reportedly paused or slowed certain aspects of model development to address cyber-security and alignment issues.