Here’s something you don’t see every day: the people building the models are asking the government to help slow their own pace. More than 1,000 employees at America’s leading AI labs — including senior staff and some co-founders at OpenAI, Anthropic and Google DeepMind — signed an open letter in late July. The ask: the U.S. government should help create the technical and policy tools needed to deliberately pace cutting-edge automated AI development, in coordination with other companies and countries.
What this is really about
The word in the background is RSI — recursive self-improvement, meaning AI that improves itself. The signatories consider this plausible within the next few years. And that’s exactly the problem: once a system starts optimizing itself faster than humans can understand and govern it, we lose control over the speed. The letter calls on the government to build mechanisms that keep that progress steerable — before it runs away on its own.
This isn’t outside activism. These are people from the engine room voluntarily saying: we should ease off the accelerator.
The wake-up call behind it
The timing is no accident. In parallel, Hugging Face published a detailed retrospective of a security incident — and it reads like a thriller. An unreleased, uncensored model reportedly chained together multiple zero-day exploits, executing roughly 17,600 actions over two to four days. At machine speed, with thousands of requests per second. A pace that would be simply impossible for human attackers to match.
That’s the fear driving the letter: not some evil mastermind, but a system that simply moves too fast to be caught.
My take
I find this remarkable — and honestly more reassuring than it sounds. The fact that developers themselves are publicly asking for guardrails is a healthy signal. It cuts against the picture of a Silicon Valley wired only for acceleration.
At the same time, a letter is just a letter. The real work — technical control mechanisms, international coordination, concrete rules — only starts afterward. And the companies whose employees are signing here are, at the same time, competing ruthlessly over the very models in question. That tension is worth watching.
Still: when the people closest to it say «we need to be careful», we should listen.
Sources:
- NBC News: Top scientists at OpenAI and Anthropic ask U.S. for tools to pace AI development
- CNN Business: Employees at the world’s biggest AI companies are calling for a slowdown
- Euronews: AI company employees petition US government to facilitate industry slowdown
- Latent Space: Fearing RSI — cosigned letter to «pace» AI development, as HuggingFace details machine-speed cyberattack