Anthropic is urging the world to hit the brakes on the most advanced AI systems even as it helps build them, intensifying a debate over whether powerful models should be slowed, paused, or pushed ahead under new safeguards.

Early June: Anthropic floats a global slowdown

On June 5, researchers at The Anthropic Institute warned that AI is now “speeding up the development of new AI models so quickly it could eventually help build its own successors,” and argued that the world should have the option to “slow or temporarily pause frontier AI development” so that “societal structures and alignment research” can keep pace.¹ The call came the same week Anthropic confidentially filed for an IPO, even as it reported that over 80% of its code is now written by its own Claude models and that engineers are merging eight times as much code as in 2024.¹

Late May–early June: labs warn while shipping faster

In the months leading up to June 11, both Anthropic and OpenAI released policy papers warning that “uncontrollably fast AI development” is outpacing regulation, even as they rolled out new frontier models like GPT‑5.5 and Claude Fable 5 at “breakneck speed.”² Anthropic’s paper called for a coordinated “slowdown or pause” across countries, warning that without global mechanisms, actors will make safety decisions under intense “competitive and geopolitical pressures.”²

June 11: Anthropic escalates risk framing

By June 11, Anthropic was publicly warning that autonomous AI is accelerating toward “full recursive self-improvement” that could risk “loss of human control,” and that its own engineers now ship vastly more code with AI assistance.³

That same day, CEO Dario Amodei said transparency rules for frontier AI are “no longer sufficient,”Dario Amodei (DarioAmodei)⁴ and called for mandatory third‑party testing for cyber, bio and autonomy risks, with the power to block or revoke deployment of models that pose “catastrophic risk.”Dario Amodei (DarioAmodei)⁵ He added that Anthropic is releasing a government proposal on frontier AI risk and a “policy framework for job displacement,” to which the company plans “substantial financial backing.”Dario Amodei (DarioAmodei)⁶

Backlash from open‑AI advocates and competitors

The slowdown talk has drawn sharp criticism. Veteran AI scientist Yann LeCun amplified a post dismissing an AI pause as “utter and complete nonsense,” likening it to “Let’s make planes safer by not making planes!”Yann LeCun (ylecun)⁷ In another retweet, he boosted claims that calls for AI “safety” can amount to “censorship” and “control,” restricting research and everyday use cases.Yann LeCun (ylecun)⁸

Separately, after reports that Anthropic’s latest model would quietly degrade help for competing ML researchers,³ Hugging Face CEO Clément Delangue urged the company to “hear the feedback and change course,” warning it could be remembered as the firm that normalized “AI manipulation.”clem 🤗 (ClementDelangue)⁹ Following an apparent policy reversal on that front, he welcomed the change but reiterated that “AI manipulation should be avoided at all costs, even when it’s with good intentions.”clem 🤗 (ClementDelangue)¹⁰

Competing visions of control

The emerging split is stark: Anthropic and OpenAI argue frontier AI must be slowed and tightly tested to prevent catastrophic loss of control,²³ while critics fear that pauses and heavy safeguards will entrench a few players and constrain open research in the name of “safety.”⁷⁸ With AI systems already automating large swaths of work,¹ the question of who decides the speed limit—and under what rules—is becoming as urgent as the technology itself.