• Rising capabilities and early warnings
• Claude’s role and internal impacts
• Calls for a coordinated slowdown
• Wider industry and expert context

Anthropic Calls for a Slowdown or Pause in Frontier AI Development Anthropic, one of the leading frontier AI labs, is urging its own industry to consider hitting the brakes just as its models become powerful enough to transform how AI itself is built. The company argues that without a coordinated slowdown, society and safety research may not keep pace with accelerating capabilities.

§Rising capabilities and early warnings

In early June, Anthropic publicly warned that AI systems are beginning to change how new AI is developed, with frontier models rapidly improving coding, debugging, and research—laying the groundwork for a self-reinforcing feedback loop in which AI builds more advanced successors. The company highlighted the prospect of “recursive self-improvement,” where AI could autonomously design, build, and train more capable versions of itself.

Anthropic stressed that such systems do not exist yet and are not guaranteed, but could arrive “sooner than most institutions are prepared for.” In a blog post, it framed RSI as an AI system “capable of fully autonomously designing and developing its own successor.”

§Claude’s role and internal impacts

New data from Anthropic’s research arm shows its flagship model, Claude, already writes more than 80% of the code merged into the company’s production codebase, up from low single digits in early 2025. Engineers now ship around eight times more code per day than in 2024, and one engineer reported not writing a line of code in five months because the model does it instead.

This surge in capability is reshaping work. One Anthropic employee captured the psychological whiplash: “On days where everything works well, I can’t help but think nothing I do matters… But then there are days where everything breaks and I don’t understand why and I realize I have no idea what I’ve been up to anymore.”

§Calls for a coordinated slowdown

Against this backdrop, Anthropic’s Institute argues that leading frontier labs “may need to slow down” so societal structures and alignment research can catch up, and that a unilateral pause is insufficient without global coordination. A recent paper, “When AI builds itself,” calls for a verifiable global pause mechanism for frontier AI development.

Executives emphasize the need to “figure out the tools to validate and verify that the stuff being done by these AI systems is correct and is aligned with human intentions.” They also want lawmakers briefed on recursive self-improvement before it becomes a mainstream political issue.

§Wider industry and expert context

The warning lands in a tech sector already boasting that large shares of code are AI-generated—Google has said 75% of its code is now written by AI, amid AI-linked layoffs and restructurings. At the same time, some AI researchers, like Meta’s Yann LeCun, signal skepticism toward more speculative fears, amplifying arguments that even highly capable systems like AlphaFold or image and video generators are not evidence of machine consciousness.

Anthropic maintains that while recursive self-improvement is not inevitable, failing to prepare for it could leave regulators, companies, and workers scrambling to adapt.

Continue reading https://foxvector.com/stories/019e9a10-a60c-3b3f-73be-27662249d1a7