Anthropic's call to slow the pace of frontier AI development won public backing from OpenAI, SpaceX and Google DeepMind's leaders, but Meta's Mark Zuckerberg is going his own way, and the industry cannot even agree on how a slowdown would actually work.
Anthropic Chief Executive Dario Amodei published an essay this month calling on the artificial intelligence industry to "pace the frontier," arguing that the most advanced AI systems are progressing faster than humans can safely monitor and control. OpenAI's Sam Altman, SpaceX's Elon Musk and Google DeepMind's Demis Hassabis each voiced support for the idea, a notable moment of alignment among executives who typically compete aggressively on everything from model architecture to talent. Meta Chief Executive Mark Zuckerberg took the opposite position, writing that "every lab has the responsibility and incentive to move at the pace required to train its models safely," and noting that Meta delayed the release of its own Muse AI agent for several months over safety concerns without calling on competitors to do the same.
The public alignment among Amodei, Altman, Musk and Hassabis followed weeks of rising unease inside the industry. A researcher who resigned from Anthropic this month, Jacob Coxon, wrote publicly that OpenAI and Anthropic are "gambling with our lives" by moving too fast. Anthropic's own head of alignment science, Evan Hubinger, has said he sees a "greater than 10% chance" of AI eliminating humanity within the decade. OpenAI's policy chief, Chris Lehane, said the company has been collaborating with Anthropic and Google DeepMind on safety issues for several weeks, suggesting the public essay reflects conversations already under way behind the scenes.
Beneath the shared language of caution, the industry has not settled on what pacing the frontier would actually mean in practice. Ritwik Gupta, a computer science and AI professor at the University of Maryland, argues that any slowdown depends on cooperation that may not survive today's competitive dynamics, since a lab that holds back risks ceding ground to a rival willing to keep pushing. He is also skeptical that limiting computing power used in training alone would meaningfully change the pace of progress, since much of today's advancement comes from software refinements and expanded use of existing models rather than from bigger training runs. Ben Goertzel, a longtime AI researcher who helped popularize the term "artificial general intelligence," goes further, arguing there is "no coherent, rational slowdown plan to be considered" once the line between AI development and ordinary software development becomes as blurred as it has.
A.J. Bhadelia, who oversees public policy for Cohere in the Americas, said some of the recent discussion around AI risk has strayed into "science fiction" territory, pointing specifically to Hubinger's 10% figure as lacking any real scientific grounding. Her view is that the industry needs an evidence-based conversation about the actual types and severity of AI risk before it settles on policy, since a consumer chatbot and an enterprise infrastructure platform plausibly require very different guardrails.
One proposal has gained traction across the disagreement: placing independent evaluators inside frontier labs with employee-level access to monitor practices, an idea Amodei raised directly in his essay. A coalition called the AI Evaluator Forum published a letter with more than 100 signatories this month outlining a framework for how such evaluators should operate. Nat Purser, director of U.S. policy at the AI Verification and Evaluation Research Institute, said the central open question is independence, specifically whether evaluators assigned by the labs themselves can speak freely without fear of retaliation, or whether a licensed pool of auditors assigned by regulators would better preserve that independence. For investors in AI-exposed companies, the unresolved questions, over cooperation, over what a slowdown would even restrict, and over who watches the watchers, mean the policy outcome remains genuinely unsettled even as the industry's most prominent leaders now at least agree the conversation is overdue.