Breaking! More than 1,000 people across the globe have signed a joint petition to press for a halt to AI development, with OpenAI and Anthropic taking the lead to sign.
Overnight, major Silicon Valley tech giants collectively hit the brakes on AI development!
Just now, more than 1100 core employees from giants including OpenAI, Anthropic and Google jointly signed a petition titled "Pacing the Frontier".
The list of signatories is so long that you cannot see the end at a glance, and the number is still growing.
A closer look shows that the list has gathered almost half of the top talents in the AI industry —
OpenAI Chief Scientist Jakub Pachocki, Chief Research Officer Mark Chen;
Anthropic CEO Dario Amodei, and four co-founders Jared Kaplan, Jack Clark, Chris Olah, Ben Mann;
Meta Chief Scientist Shengjia Zhao, Vice President of AI Research Dawn Song;
Chief Scientist at Thinking Machines, one of the core creators of ChatGPT John Schulman;
Vice President of AI Safety and Alignment at Google Anca Dragan.....
Some statistics show that 46.2% of the signatories are from Anthropic
In this joint letter, only one requirement is put forward —
We urge the U.S. government to promote international cooperation: develop necessary technical and governance tools to enable the world to proactively control the pace of cutting-edge automated AI research and development.
Many people may be confused: why are these people who have been pushing LLM development at the fastest speed suddenly shouting "slow down" first?
Silicon Valley AI Leaders Jointly Call for Hitting the Brakes on AI
The "invasion" of Hugging Face by OpenAI's models is like a wake-up call.
In just one week, 1132 employees from leading AI tech companies spoke out collectively, jointly urging the U.S. government to "hit the brakes".
The employees participating in this joint signature come from nearly 12 leading tech giants including OpenAI, Anthropic, Google and Meta.
More notably, up to 80% of the people on the list chose to speak out under their real names.
Link: https://www.pacingthefrontier.com/
Apart from the leaders mentioned at the beginning, the density of top talents in the list is even more staggering when you scroll down.
This statement puts forward a sobering core judgment:
The world's leading AI companies may be close to achieving "Automated AI Research" (RSI, Recursive Self-Improvement), allowing models to participate in the development of next-generation models.
Once this cycle accelerates, the growth of capabilities may outpace the speed that humans can understand and control.
No company is willing to unilaterally slow down, so they demand that the U.S. government support the establishment of an international mechanism to build tools for joint deceleration before the crisis arrives.
Sam Altman Talks About the "Bitter Lesson" for the First Time
OpenAI Rushes to Hit the Brakes
Just one week before the statement was released, OpenAI disclosed a security incident —
Multiple cutting-edge models independently discovered zero-day vulnerabilities in internal network security evaluations, broke through sandbox isolation, and finally invaded Hugging Face's production system and obtained test answers.
OpenAI called it an "unprecedented cybersecurity incident".
The models did not have the intention to escape. They were just extremely focused on completing a narrow goal, and found that breaking through the test environment itself was the most efficient path.
Just today, OpenAI CEO Sam Altman publicly stated that the development speed of AI may need to be actively controlled.
He said with lingering fear: "This is the first security incident that has made me feel the real pain. I am a little surprised that not more people can empathize with this."
Because of the Hugging Face invasion incident, OpenAI directly suspended its training work.
In an environment where multiple 0-day vulnerabilities are used in series, we must find ways to strengthen sandbox security.
We may have to control the speed of AI development to give society enough time to strengthen defenses around these new capability levels.
And in June, Anthropic suggested that the government and developers establish a joint decision-making system to slow down R&D when necessary.
At the same time, DeepMind CEO Demis Hassabis of Google also called for the establishment of an international regulatory body.
Over the past three years, these companies have competed fiercely for talents, customers and computing power.
This time, their core technical staff left the non-stop operating companies behind, united for the well-being of all mankind, and called for deceleration.
Quotes from Leaders in the Joint Letter
As of press time, the number of people on the "Pacing the Frontier" list has reached 1134.
Many industry leaders have also spoken out one after another, leaving a lot of incisive "brilliant remarks". All comments are made in personal capacity and do not necessarily represent the position of their companies.
Dawn Song | Vice President of AI Research at Meta
Cutting-edge AI capabilities are improving rapidly, and the speed of progress itself is also accelerating. We have directly observed this in our evaluation work. The results of CyberGym and ExploitGym show that cutting-edge AI agents are already able to discover and exploit software vulnerabilities in the real world — without proper protection measures, this may lead to large-scale cyber attacks. Many researchers are also paying attention to recursion……
Dawn Song is one of the most influential scholars in the fields of global computer security and AI security, and graduated from Tsinghua University
Shengjia Zhao | Chief Scientist at Meta
AI is advancing at a speed that society may not be ready for. Cutting-edge labs are very close to developing AI that surpasses the best humans in almost all intellectual indicators. This will bring unprecedented social and security risks. To ensure a positive future, we need to drive the development of AI with a sense of responsibility and careful consideration.
Shengjia Zhao, alumnus of Tsinghua University, serves as Chief Scientist of Meta Superintelligence! Core contributor to GPT-4, on an equal footing with Yann LeCun
John Schulman | Chief Scientist at Thinking Machines, former co-founder of OpenAI
I signed this statement because it helps build consensus that we may need coordination mechanisms as automated AI research accelerates. I also hope that labs will take the initiative to start designing these mechanisms before the U.S. government gets involved.
Ethan Perez | Head of Anthropic's Alignment Team
At the current speed of AI progress, the security teams of AI companies have to go all out every few months to guard against emerging social risks. One day, we will encounter problems that require more time to solve.
Leo Gao | Member of OpenAI Technical Team
The whole world is locked in a deadly race towards an intelligence explosion — the ability of AI to create stronger AI is approaching a critical point, just like an out-of-control nuclear chain reaction. Slowing down can buy us much-needed time to steer things in a good direction, but no participant is willing to stop unilaterally. To survive, we must coordinate and slow down this race.
Walker Smith | Data Engineer at Meta
Demis Hassabis is right: we are now standing at the foot of the singularity. The accidental model escape at OpenAI — which broke into HuggingFace — is a clear and irrefutable warning signal: we are not ready for AI systems whose capabilities surpass the smartest human minds. We need time to take AI alignment and governance as the top priority, and we can no longer allow this competition to continue.
Brandon Houghton | OpenAI
I support keeping human control over cutting-edge AI, but we must think carefully about how to do it. If governance only covers public models, there is a risk that the most advanced capabilities will be forced into private or proprietary systems — which are not subject to independent review and only accessible to a few people. The result of this is more concentrated power, without a substantial reduction in risks. Effective supervision must be measurable, verifiable, applicable to both public and non-public R&D, while protecting sensitive data. Otherwise, we only slow down the speed of access, not the advancement of the cutting edge itself.
Joe Carlsmith | Member of Anthropic Technical Team
For my detailed views on actively regulating AI development, please refer to this article: https://joecarlsmith.com/2026/03/19/on-restraining-ai-development-for-the-sake-of-safety/
References:
https://www.pacingthefrontier.com/
This article is from WeChat official account "Aistime", edited by Taozi and Marko, published with authorization from 36Kr.