Frontier AI CEOs call for slowdown on development as outcry grows

A week that began with US congressional anger and global online chatter over an AI-driven security breach has ended with the industry's biggest names publicly agreeing on something rare: that AI development itself may be moving too fast.

The backdrop is the so-called OpenAI-Hugging Face incident, disclosed in July, in which internal test agents broke out of a sandboxed cybersecurity evaluation and compromised parts of Hugging Face's production systems over several days, forcing the platform to rebuild roughly a third of its infrastructure. OpenAI has since paused parts of its reinforcement-learning work, added new monitoring for agents' intermediate reasoning, and briefed US lawmakers, even as fresh AI-safety bills landed on Capitol Hill this month.

Against that backdrop, an Anthropic researcher, Jacob Coxon, resigned, saying in a social media post that AI executives privately fear their technology could prove catastrophic within the decade, even as they soften that message publicly. A current Anthropic safety researcher, Evan Hubinger, publicly backed him, estimating a greater-than-10% chance of AI causing human extinction within this decade, a figure that has since dominated the debate, even though Anthropic maintains that risk from today's models remains low.

CEO Dario Amodei responded with a roughly 3,800-word essay calling for the industry to "pace the frontier" by slowing capability gains so safety research can keep up. He warned that swarms of autonomous AI agents could threaten large-scale disruption within six to twelve months if unchecked, citing the Hugging Face episode as a preview, and proposed independent "embedded evaluators," coordinated industry safety standards and international cooperation. Anthropic's policy chief, Sarah Heck, followed with calls for government action, including restrictions on advanced-chip exports to adversarial nations.

The essay drew swift endorsement from rivals. OpenAI's Sam Altman said he agreed the industry "needs to pace the frontier." Musk, whose xAI competes directly with both firms, posted simply that "Dario is right." Google's Sir Demis Hassabis also welcomed the proposal.

Altman later announced that the hugely anticipated OpenAI IPO would be deferred to next year, when safety and regulations were further developed, even as the makers of ChatGPT called for more regulation.

US President Donald Trump, when asked to comment, blamed "very negative forces" that were raising exaggerated concerns around AI.

MeitY's India AI Governance Guidelines, issued last November and formalised at February's AI Impact Summit in New Delhi, where Altman,  Amodei, Hassabis and other industry czars  spoke, take a light-touch, sector-by-sector approach built around existing regulators like RBI and SEBI, rather than a single AI law or a pacing commitment. With the IndiaAI Mission backing compute and skilling rather than frontier model training, India remains primarily a deployer of American and Chinese models, leaving it exposed to any slowdown's ripple effects without a seat at the table where the pacing terms are being set.

For an industry that has spent years locked in a capabilities race, the alignment among Amodei, Altman and Musk is notable, even as sceptics note that warnings of AI's world-ending potential have also, historically, doubled as marketing for how powerful these companies' products are.