Anthropic CEO Calls For Slower AI Development Amid Growing Misuse Concerns

Anthropic CEO Dario Amodei has urged AI companies to slow down the pace of model development, citing the need for more time to manage risks associated with increasingly capable systems. In a recent X post, Amodei outlined a three-step framework aimed at balancing AI progress with addressing potential threats. Earlier this year, Anthropic had already advocated for a slowdown in AI development, and Amodei's latest proposal aligns with these concerns.

The CEO argues that recent incidents of AI misuse and the emergence of systems capable of recursive self-improvement have heightened the urgency for such measures.

The first step in Amodei's framework involves frontier AI companies granting independent third-party evaluators ongoing access to their operations. These evaluators would verify compliance with safety standards, assess whether model training aligns with efforts to slow development, and report any incidents. The second step involves AI companies collaborating with governments to establish common safety standards, aiming to curb unchecked AI development and ensure a consistent approach to risk management.

Finally, Amodei emphasizes the importance of international cooperation, urging coordination between the US, democratic countries, and authoritarian governments to ensure consistent application of AI safety measures worldwide.

Anthropic has committed to the first step, and OpenAI CEO Sam Altman expressed agreement with the proposal, stating that the company would also provide independent evaluators with employee-like access. Elon Musk, CEO of xAI, also echoed Amodei's call for a slower AI development pace. The call for slower AI development follows Anthropic's release of a threat intelligence report highlighting the misuse of its Claude AI models, including for cyberattacks, weapons development, surveillance, fraud, and biological research.

The report also uncovered an alleged commercial influence operation targeting Malaysian voters, which Anthropic said used Claude to profile constituents, manage fake social media accounts, and generate political content through a fabricated news outlet. The Malaysian Communications and Multimedia Commission (MCMC) has been reviewing the allegations and denies involvement in the described activities.