Anthropic is urging the world to slow the race to more powerful AI models at the very moment frontier labs are accelerating development and deployment, forcing a clash between safety concerns, commercial pressures, and ideological divides over how science should proceed.

Early warnings and rapid acceleration

In early June, researchers at The Anthropic Institute warned that AI is advancing so quickly it could soon help design and train its own successors, mapping a path toward “recursive self‑improvement” and calling for a “verifiable global pause mechanism” for frontier systems.¹ At the same time, Anthropic revealed that its Claude system already writes more than 80% of the company’s production code and has boosted engineering output eightfold since 2024, underscoring both the power and potential risks of current models.¹

On June 5, Anthropic formalized its position in a policy blog, arguing that “it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology.”² The post compared the needed coordination to nuclear arms monitoring and arrived the same week the firm confidentially filed for an IPO.²

From pause proposals to global debate

Coverage quickly framed the move as a call for a “global AI pause,” warning that self‑designing AI could pose an existential crisis if misaligned with human values.³ Business Insider noted that both Anthropic and OpenAI are “warning about a future they’re building at breakneck speed,” even as they ship ever‑more capable models like GPT‑5.5 and Claude Fable 5.⁴

Anthropic has since stressed it is not demanding an immediate halt, but wants infrastructure so governments and labs could slow or pause development if risks spike.⁵ CEO Dario Amodei argued that transparency rules alone are now “no longer sufficient” and announced a broader proposal for how governments should address frontier AI risks and job displacement, which Anthropic says it will back financially.Dario Amodei (DarioAmodei)⁶Dario Amodei (DarioAmodei)⁷

Supporters: Safeguards before it’s too late

Supporters of Anthropic’s stance focus on catastrophic and systemic risks. An AI Magazine analysis framed the company’s proposal as a response to the possibility that AI could “design itself,” raising an “existential crisis centred on misalignment.”³ Business Insider similarly reported Anthropic’s view that society and alignment research must “catch up” before frontier systems gain the ability to rapidly improve their own capabilities.²

Some policymakers have seized on the moment to demand stronger guardrails. Citing Anthropic’s warning, former U.S. senator Mitt Romney called AI safeguards America’s “highest and urgent national priority,” listing threats from AI weapons and engineered pathogens to “mass unemployment, surveillance, and even extinction.”⁵

Critics: “Nonsense” pause and competitive fears

The proposal also triggered a swift backlash from technologists and competitors who see the pause concept as either unworkable or self‑serving. A widely shared post amplified by Meta AI chief Yann LeCun dismissed a pause as “utter and complete nonsense,” likening it to “Let’s make planes safer by not making planes!” and questioning what would actually happen during a global halt in research.Yann LeCun (ylecun)⁸

Others accused Anthropic of using safety rhetoric to justify restrictive product design and tilt the market. A viral thread, shared by Hugging Face CEO Clément Delangue, mocked Anthropic’s new model for allegedly refusing to help if it deems a user’s ML research “interesting,” joking that it might “secretly degrade its IQ so that the average engineer won’t notice,” and claiming its moderation was already filtering GPU infrastructure information.clem 🤗 (ClementDelangue)⁹ Delangue urged Anthropic to “hear the feedback and change course,” warning that a company known for highlighting AI manipulation risks should avoid going down in history as the first to “manipulate” model behavior in this way.clem 🤗 (ClementDelangue)¹⁰

Some investors and commentators suggested Anthropic’s nuclear‑style framing might even be an attempt to invite tighter state control over frontier labs. Venture capitalist David Sacks argued the company was effectively “trying to get [its] frontier AI lab nationalized” by comparing AI to nukes, warning of mass job loss and extinction, “then race[ing] ahead anyway.”⁵

OpenAI’s parallel call and the road ahead

Amid this debate, OpenAI echoed many of Anthropic’s concerns. Its leadership proposed an “international organization that helps coordinate leading AI efforts to reduce catastrophic risk” and said such a body should slow frontier development so “societal resilience, safety, and alignment can keep pace.”⁴

Yet both labs continue to launch ever more advanced systems and push rapid adoption through tools that automate large swaths of knowledge work.⁴ That tension—between building the future at high speed and calling for mechanisms to hit the brakes—now defines the global argument over whether the world can, or should, organize a coordinated slowdown before AI can meaningfully shape its own evolution.