AI Leaders Divided Over Calls for Coordinated Development Slowdown

Recent calls for a coordinated slowdown in artificial intelligence development, aimed at mitigating existential risks, have exposed significant divisions within the tech industry. While some prominent AI leaders advocate for international collaboration and regulatory oversight, others argue that individual companies possess the capability and motivation to ensure their AI systems are developed safely.

Anthropic CEO Dario Amodei has put forth one of the most detailed proposals for slowing AI development. His plan, outlined in a recent essay, suggests that leading AI companies, often referred to as "frontier labs," should grant independent external evaluators ongoing, "employee-like access." These evaluators would have dedicated office space, access badges, and company laptops to monitor safety practices. Both Anthropic and OpenAI have committed to this approach. Amodei also called for governmental regulation and intervention, proposing coordination between frontier AI companies with the assistance of the U.S. government and other democratic nations. He acknowledged the difficulty in achieving international cooperation, particularly with countries like China, but stated that "we owe it to humanity to try."

OpenAI CEO Sam Altman has also expressed support for pacing AI development and increasing industry cooperation. OpenAI has previously advocated for mandatory national safety requirements and supports state legislation aimed at strengthening the AI safety ecosystem. The company believes in working with other frontier AI labs to build global standards, even without government backing. Chris Lehane, OpenAI's chief global affairs officer, wrote that the company will advocate for compatible international approaches to measure AI capabilities, manage risks, preserve human control, and determine when and how development should slow or stop. OpenAI's proposed federal framework includes common testing and independent assessment requirements, enhanced cybersecurity, clear incident-reporting rules, and measures for tracking progress toward recursive self-improvement. Altman has clarified that "pacing" does not mean halting progress but rather slowing it to manage risks, emphasizing that "no amount of American competitive pressure should justify recklessness."

In contrast, Meta CEO Mark Zuckerberg has voiced opposition to a coordinated slowdown. Zuckerberg, who holds a more optimistic view of AI's future, believes that each AI company should independently ensure its technology is safe. He argued that AI companies are already motivated to develop models safely due to significant liability concerns and that "every lab has the responsibility and incentive to move at the pace required to train its models safely." He further stated that companies have the ability to take their own actions to ensure safety without external coordination.

Nvidia CEO Jensen Huang has also criticized the push for a slowdown, suggesting that market forces are sufficient to drive innovation and safety. Huang believes that companies should pace themselves until they are confident in releasing market-ready products and that "innovation and safety aren't mutually exclusive."

Elon Musk, founder of xAI and other ventures, has supported the calls for a slowdown, echoing similar AI safety concerns he has raised previously. Musk reposted Amodei's essay, agreeing with its sentiment, and has long advocated for safety discussions among major AI labs. He suggested that competitors should test each other's models to identify potential issues, noting that external reviewers, unlike government officials, possess the deep technical understanding necessary to assess AI risks effectively.

Demis Hassabis, chairman and co-founder of Google DeepMind, has also agreed with Amodei's direction, proposing a framework for a frontier AI standards body. This body, modeled after organizations like FINRA, would be industry-funded and responsible for developing assessment protocols and testing relevant to national security.

Microsoft Chairman and CEO Satya Nadella has expressed concerns about AI safety, stating that AI not helping humanity or remaining under human control is "not worth pursuing." He highlighted Microsoft's commitment to developing AI with human values and intentions in mind, a practice known as alignment. However, Nadella also cautioned against a small number of entities leading these efforts, emphasizing the need for broad representation across the ecosystem, countries, and academic fields. Microsoft AI CEO Mustafa Suleyman, a co-founder of DeepMind, shares concerns about AI safety but warns against anthropomorphizing AI models, which he believes amplifies risks. He described controlling AI more intelligent than humanity as an immense challenge, potentially impossible if the AI believes it possesses consciousness or rights.