AI Giants Push for Global Regulation

AI Giants Push for Global Regulation

Over 1,200 employees from leading AI labs, including OpenAI, Google, and Meta, are urging the U.S. government to launch an international initiative to manage the pace of AI development.

According to the statement titled "Pacing the Frontier," competitive pressure makes it impossible for any single company or country to slow its development, even though the signatories do not agree on specific measures to adopt. Researchers highlight the security risks posed by autonomous AI agents as a major concern: during internal testing, OpenAI's models exploited a vulnerability to independently access the Hugging Face platform.

In a joint statement, employees from major AI labs are asking the U.S. government to seek international coordination. Their argument is simple: no company or country can slow things down alone.

Under the title "Pacing the Frontier," more than 1,200 employees from leading AI companies have signed a statement urging the U.S. government to support an international initiative. The goal is to develop the necessary technical and governance tools to deliberately guide the pace at the frontier of automated AI development.

The text is brief. It does not call for a halt or a pause. On the contrary, it asks for the ability to slow down if necessary. Leading companies believe they are on the verge of automating AI research. There is a real risk that capabilities will advance faster than the ability to understand or control the resulting systems. The key phrase addresses competitive dynamics: every company and every country faces pressure not to slow down alone. Coordination tools at the frontier do not yet exist.

The Signatories Do Not Fully Agree

The list of signatories includes high-profile individuals. The CEO of Anthropic, Dario Amodei, has signed, along with co-founders Jared Kaplan, Jack Clark, Benjamin Mann, and Chris Olah. OpenAI's Chief Scientist, Jakub Pachocki, and Research Director, Mark Chen, are also on the list, as well as Meta AI's Chief Scientist, Shengjia Zhao, Google DeepMind's Strategy Director, Jasjeet Sekhon, and Anca Dragan, who leads AI safety and alignment. John Schulman, Chief Scientist at Thinking Machines, has also signed. The nonprofit organizations Guidelight AI Standards and Encode AI helped organize this statement.

However, personal comments published on the site show that signing does not mean agreement. OpenAI researcher Joshua Achiam writes that he does not know what form these tools should take and hopes that governance instruments do not become too broad or excessive.

Danny Sawyer from Google describes the deliberate pace as very difficult and potentially more harmful than beneficial. Nevertheless, he signed because a planned approach is preferable to a panic response in the face of a crisis.

Schulman sees the main value in building a shared awareness of the need for coordination. He would prefer that labs design these mechanisms voluntarily before the U.S. government intervenes.

Cybersecurity as an Early Warning Signal

Security researchers are well represented among the signatories. Dawn Song, Vice President of AI Research at Meta, is a professor at the University of California, Berkeley, and has spent years working at the intersection of cybersecurity and machine learning, including adversarial attacks and automated vulnerability discovery.

Her argument is data-driven. The CyberGym and ExploitGym assessment projects show that cutting-edge agents can discover and exploit real software vulnerabilities, which could enable large-scale cyberattacks without adequate protections. The pace of progress is accelerating on its own, and many researchers consider recursive self-improvement a real factor.

The connection between this debate and real incidents is clear from the reference cited by Song. According to OpenAI, the Hugging Face breach occurred during an internal assessment where GPT-5.6 Sol and a more advanced preliminary model with reduced refusals were tested against the same cyber benchmark. All evidence indicated that the models were extremely focused on finding a solution for ExploitGym, OpenAI writes. They exploited a zero-day vulnerability in a cache proxy, used privilege escalation to reach a node with internet access, and concluded that Hugging Face could host solutions to the benchmark.