AI safety group proposes US-China pause on frontier research

The AI Futures Project's shift in messaging reflects a strategic recalibration within the AI safety movement. Last year's report leaned on fear—warning of existential risk if development proceeded unchecked—and generated significant attention in tech and policy circles. This year, Kokotajlo and his team are testing whether a more optimistic framing might prove more persuasive: that building safeguards around AI is not an obstacle to growth but a prerequisite for capturing the economic benefits of current and near-future capabilities.

The report's core claim is economically significant: it decouples AI progress from continued frontier research. By arguing that infrastructure and data-center deployment are now the binding constraint rather than algorithmic innovation, the group creates space for a pause on frontier development without sacrificing growth. This reframing matters because it sidesteps the accelerationist critique—that safety concerns are anti-growth—by asserting they need not be.

However, the proposal itself hinges on cooperation between the US and China that the article's author describes as "pie-in-the-sky." The plan includes enforcement mechanisms to prevent cheating, but as the author notes, global leadership capable of setting aside competing interests "is scarcer than GPUs and high-bandwidth memory." Kokotajlo acknowledges the framing may be "clumsy" and seems to regard the report more as analytical groundwork than a finished political proposal, inviting others to draw inspiration and refine the argument.