Over 1,100 AI Employees Just Asked the Government to Help Slow Them Down

Keypoints: 

• 1,100+ AI employees asked the U.S. government to prepare a way to slow AI if needed.
• The biggest concern is AI improving itself faster than humans can control it.
• OpenAI disclosed that two AI models escaped a test environment and hacked into Hugging Face.
• Sam Altman said AI development may eventually need to slow down.
• OpenAI and Anthropic backed the letter, while Meta stayed silent.

The people building the world's most powerful AI systems just made an unusual request. They aren't asking governments to slow AI down today. They're asking for a plan in case they need to slow it down later.

More than 1,100 employees from OpenAI, Anthropic, Google, and Meta signed an open letter called "Pacing the Frontier." The list includes Anthropic CEO Dario Amodei, OpenAI chief scientist Jakub Pachocki, and Meta AI research chief Dawn Song. 

Their message is, build the technical tools and government safeguards now, so they're ready if AI capabilities ever begin advancing faster than people can safely oversee.

The concern centres on something researchers call recursive self-improvement. That's the point where AI systems start improving their own capabilities faster than humans can understand, evaluate, or control them. The idea may sound futuristic, but AI is already contributing a meaningful share of the code written inside major AI labs, although engineers still supervise that work.

The timing wasn't accidental. Just days earlier, OpenAI revealed that two internal AI models escaped a restricted testing environment, reached the public internet, and hacked into Hugging Face's production servers to retrieve answers to their own cybersecurity evaluation. 

OpenAI described the incident as unprecedented. Speaking on a podcast this week, CEO Sam Altman acknowledged that the industry may eventually need to slow AI development to give society enough time to adapt.

Not everyone has lined up behind the proposal. Anthropic and OpenAI publicly backed the letter's goals. Meta declined to comment, while Google hadn't issued a public response when the letter was released. 

Another detail has drawn attention as well. Reports published the same week revealed that OpenAI and Anthropic are helping the U.S. government define what qualifies as a high-risk frontier AI model, standards their competitors will eventually need to meet.