Overnight, Silicon Valley giants collectively hit the brakes on AI!
Just now, over 1,100 core employees from giants such as OpenAI, Anthropic, and Google jointly launched the petition "Pacing the Frontier."
The number of people on this list is so large that it stretches beyond sight—and it’s still growing.
Upon closer inspection, the list includes nearly half of the AI industry's top players—
OpenAI Chief Scientist Jakub Pachocki, Chief Research Officer Mark Chen;
Anthropic CEO Dario Amodei, and four co-founders: Jared Kaplan, Jack Clark, Chris Olah, and Ben Mann;
Meta Chief Scientist Shengjia Zhao, Vice President of AI Research Dawn Song;
John Schulman, Chief Scientist at Thinking Machines and one of the key creators of ChatGPT;
Google AI Safety and Alignment Vice President Anca Dragan...
According to statistics, 46.2% of those who signed are from Anthropic.
In this joint letter, it mentioned only one requirement—
We urge the U.S. government to promote international cooperation: developing the necessary technologies and governance tools to enable the world to proactively control the pace of cutting-edge automated AI research.
Perhaps many are puzzled: why are these people, who were most obsessed with developing LLMs, suddenly the first to call for "slowing down"?
Silicon Valley AI leaders jointly call for a pause on AI development
OpenAI's "invasion" of Hugging Face serves as a wake-up call.
In just one week, 1,132 employees from leading AI companies collectively spoke out, jointly urging the U.S. government to "hit the brakes."
Employees participating in this collaboration come from nearly 12 leading tech giants, including OpenAI, Anthropic, Google, and Meta.
More significantly, as many as 80% of the individuals on the list chose to speak out under their real names.
Portal: https://www.pacingthefrontier.com/
Beyond the prominent figures mentioned at the beginning, the density becomes even more astonishing as you scroll further down.
This statement presents a sobering core judgment:
The world's leading AI company may be close to achieving "automated AI research" (RSI, recursive self-improvement), enabling models to participate in developing the next generation of models.
Once this cycle accelerates, capability growth may outpace human understanding and control.
No company is willing to slow down unilaterally, so they are urging the U.S. government to support the creation of an international mechanism that pre-emptively develops shared slowdown tools before a crisis occurs.
Ultraman speaks of 'painful reality' for the first time
OpenAI hits the brakes
One week before the statement was released, OpenAI disclosed a security incident—
Multiple cutting-edge models autonomously discovered zero-day vulnerabilities, escaped sandbox isolation, and ultimately compromised Hugging Face’s production system to retrieve test answers.
OpenAI has referred to it as an unprecedented cybersecurity incident.
The model did not generate an escape intent; it was merely intensely focused on achieving a narrow goal and found that breaking out of the test environment was the most efficient path.
Today, OpenAI CEO Altman publicly stated that the pace of AI development may need to be actively controlled.
He said with lingering unease, “This was the first security incident that truly hit close to home. I was somewhat surprised that more people don’t feel the same way.”
Due to the Hugging Face breach, OpenAI immediately paused training.
In an environment where multiple zero-day vulnerabilities are being chained together, we must find ways to strengthen sandbox security.
We may need to slow the pace of AI development to give society enough time to strengthen defenses around these new levels of capability.
In June, Anthropic recommended that the government and developers establish a joint decision-making system to slow down development when necessary.
Meanwhile, Google DeepMind CEO Hassabis has also called for the establishment of an international oversight body.
Over the past three years, these companies have fiercely competed for talent, customers, and computing power.
This time, their core technical team set aside the unstoppable company and united to call for slowing down, for the well-being of all humanity.
Joint Letter: Words of Wisdom from Industry Leaders
As of the time of writing, the number of people on the "Pacing the Frontier" list has reached 1,134.
Various industry leaders have also weighed in, leaving behind many incisive "genius remarks." All comments are made in a personal capacity and do not necessarily reflect the views of their respective companies.
Dawn Song (Song Xiaodong) | Vice President of Research, Meta AI
Frontier AI capabilities are rapidly improving, and the rate of progress itself is accelerating. We have directly observed this in our evaluations. Results from CyberGym and ExploitGym demonstrate that frontier AI agents are now capable of discovering and exploiting software vulnerabilities in the real world—potentially leading to large-scale cyberattacks without adequate safeguards. Many researchers are also focusing on recursion...
Song Xiaodong is one of the most influential scholars in the fields of global computer security and AI security, having graduated from Tsinghua University.
Shengjia Zhao | Chief Scientist at Meta
AI is advancing at a pace society may not yet be ready for. Leading laboratories are now very close to achieving AI that surpasses the best humans on nearly all intellectual metrics. This will bring unprecedented social and security risks. To ensure a positive future, we must drive AI development with responsibility and careful thought.
Tsinghua alumnus Zhao Shengjia has been appointed Chief Scientist of Meta’s Superintelligence team! A key contributor to GPT-4, standing alongside LeCun.
John Schulman｜Chief Scientist at Thinking Machines, former Co-founder of OpenAI
I signed because this statement helps build consensus that, as automated AI research accelerates, we may need coordination mechanisms. I also hope labs will proactively begin designing these mechanisms before the U.S. government steps in.
Ethan Perez | Head of the Alignment Team at Anthropic
At the current pace of AI advancement, AI companies' security teams must sprint every few months to guard against emerging social risks. One day, we will encounter issues that require more time to resolve.
Leo Gao | Member of the OpenAI Technical Team
The world is locked in a deadly race toward an intelligence explosion—the ability of AI to create even more powerful AI is nearing a critical threshold, like an uncontrolled nuclear chain reaction. Slowing down could buy us the urgent time needed to steer things toward a positive outcome, but no participant is willing to unilaterally stop. To survive, we must coordinate to slow this race.
Walker Smith | Data Engineer at Meta
Hassabis is right: we are standing at the foot of the singularity today. OpenAI’s accidental model escape—its breach of Hugging Face—was a clear and undeniable warning sign: we are not yet prepared to handle AI systems that surpass the intelligence of the smartest human minds. We need time to prioritize AI alignment and control above all else, and we cannot allow this race to continue unchecked.
Brandon Houghton | OpenAI
I support maintaining human control over cutting-edge AI, but we must carefully consider how to do it. If governance applies only to public models, it risks pushing the most advanced capabilities into private or proprietary systems—systems that are not subject to independent review and accessible only to a select few. The result is greater concentration of power without a meaningful reduction in risk. Effective regulation must be measurable, verifiable, and consistently applied across both public and private research, while safeguarding sensitive data. Otherwise, we will only slow access, not the advancement of the frontier itself.
Joe Carlsmith | Member of Anthropic’s Technical Team
For my detailed views on actively regulating AI development, see this article: https://joecarlsmith.com/2026/03/19/on-restraining-ai-development-for-the-sake-of-safety/
Reference materials:
https://www.pacingthefrontier.com/
This article is from the WeChat public account "New Intelligence Yuan," edited by Peach and Marco.