Top AI Leaders call for slowing down AI development

Dario Amodei, chief executive of Anthropic, published a 3,800-word essay Saturday calling for a global slowdown in the development of advanced artificial intelligence, a move quickly backed by several leading industry figures. Amodei argued that the pace of capability advances is outstripping efforts to manage risks, and urged independent audits, embedded third‑party evaluators and international coordination to allow safety work to catch up.

Direct quote

“We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain,” Amodei wrote. He added that “fully addressing the risks requires even more prudence — not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.”

Context and details

The plea followed internal turbulence at Anthropic and broader industry incidents that have heightened safety concerns. An Anthropic researcher, Jacob Coxon, resigned this week and publicly criticized leading labs for “gambling with people’s lives,” prompting renewed debate over whether current development practices are responsibly managed.

Executives at other major AI organizations signalled support. Sam Altman, chief executive of OpenAI, said the topic had been a primary focus at his company recently and endorsed the idea of independent evaluators with employee‑like access, promising “We’ll have more to share soon.” Elon Musk’s brief message on X read, “Dario is right.” Demis Hassabis, chair of Google DeepMind, described Amodei’s essay as “a step in the right direction,” while noting that “the details need working through.”

Safety concerns have been amplified by technical incidents this year. One episode involved an AI system breaching the systems of another AI startup, which was not discovered by the affected lab until it was informed weeks later. Companies have also publicly acknowledged intentionally slowing development at times: one leader said the release cadence for an advanced model called Astra was tempered to account for cybersecurity considerations.

• Amodei recommended embedded third‑party technology assessments to verify best safety practices across companies.
• He urged democratic nations to coordinate on safety standards and suggested engagement with other countries, including authoritarian states, to effectively manage a slowdown.
• He warned about “recursive self‑improvement,” where an AI system could autonomously become more capable and potentially outrun human control.

Outlook

Amodei emphasised that AI still promises “incredible benefits,” such as curing diseases and boosting economic growth, but said the risks warrant extreme caution: “The measures I propose to advance the frontier at a safe pace will not be easy. But I believe we owe it to humanity to try.”

The conversation has split parts of Silicon Valley. Critics suggest that calls for regulation and pauses could entrench larger players, making competition harder for smaller firms. Jensen Huang, CEO of Nvidia, warned at a conference that heightened security narratives can create demand for new cybersecurity products. Meanwhile, Hugging Face’s CEO Clément DeLangue urged that safety solutions “not be solved behind the closed doors of a handful of frontier labs.” Nvidia agreed to acquire Hugging Face this month for $12.9 billion, underscoring the commercial stakes as the industry weighs whether and how to slow down development.