OpenAI and Anthropic are urging governments to slow or coordinate “frontier” AI development—even as both labs push out new models at speed. Anthropic calls for a coordinated slowdown or pause, while OpenAI argues for an international organization to coordinate
In the last two months, two of the world’s most influential AI labs have been doing something that feels almost impossible to separate: publishing urgent warnings about the risks of rapidly advancing systems, while also rolling out brand-new models designed to push those systems further.
Anthropic and OpenAI have both framed the problem in terms of speed and governance. They argue that frontier AI development is moving uncontrollably fast, while government regulation is failing to keep up. That tension sits underneath the flurry of policy and research papers—published alongside new releases including GPT-5.5 and Claude Fable 5—and it comes as both companies prepare to make even bigger moves toward public markets.
Anthropic’s call has been the most direct. In a paper released last week. Anthropic asked for a coordinated “slowdown or pause” in frontier model development across countries so policy frameworks can catch up. The paper warned that without a global coordination mechanism. “companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures.”.
Dario Amodei, Anthropic’s CEO, repeated the concern in a blog post he published on Wednesday, saying AI is moving at a “lightning pace” while policy is “moving very slowly.”
On the other side of the same argument. OpenAI’s leadership has pushed for an institutional fix rather than an across-the-board standstill. In a Monday blog post. Sam Altman and Jakub Pachocki wrote that there needs to be the formation of an “international organization that helps coordinate leading AI efforts to reduce catastrophic risk.” They said this organization should slow frontier AI development down so “societal resilience. safety. and alignment can keep pace.”.
Those cautions have not paused the companies’ product timelines. On Tuesday, Anthropic released Claude Fable 5, described as its latest and most powerful “Mythos-class” model. The company said the model includes guardrails designed to limit harmful requests. When Claude detects requests related to topics like cybersecurity or distillation attacks. it will either route users to Anthropic’s Opus 4.8 model or steer users toward a different prompt.
OpenAI, meanwhile, released its GPT-5.5 model in late April. It called GPT-5.5 its “smartest and most intuitive” model to date, adding that it will understand what work the user is doing and can carry more of the work itself.
Both labs have also been encouraging faster adoption of their tools through free usage perks for Codex and Claude Code—tools companies can use to build AI agents that automate workflows and increase productivity.
The warnings arrive at a moment when both companies are also building toward their highly anticipated public offerings. On Monday, OpenAI said it confidentially filed its S-1 about a week after Anthropic. The public-facing timing is delicate: the companies are talking about catastrophic risk and the need for coordination while simultaneously expanding access and preparing to become publicly traded.
An analytical thread runs straight through the sequence of events already described: the same leadership teams that argue policy is too slow are also rolling out systems labeled as the latest. most powerful. and most capable versions of their models—paired with adoption incentives meant to help customers integrate these tools faster.
OpenAI and Anthropic did not respond to requests for comment from Business Insider.
OpenAI Anthropic GPT-5.5 Claude Fable 5 frontier AI AI safety regulation catastrophic risk S-1 filing Dario Amodei Sam Altman Jakub Pachocki Codex Claude Code
So basically they’re warning us while also making the next one… got it.
I don’t buy it. If they wanted a slowdown, they would stop releasing stuff. “International organization” sounds like a fancy way to delay while they catch up on patents.
Wait are they talking about pausing like right now? Because I saw a headline about GPT-5.5 and thought that already happened years ago or something. Also “Claude Fable 5”?? I’m confused if it’s even real.
This is the usual thing where companies say “govern us” then immediately sprint. Like, if policy is slow, then maybe governments should just regulate the whole tech industry instead of playing catch-up with AI labs. But knowing America… they’ll let it slide until something catastrophic happens and then act surprised.