Over 1,100 AI Insiders From OpenAI, Anthropic Demand Global Brakes on ‘Runaway’ AI Race

In an extraordinary show of unity, more than 1,100 employees and core researchers from the world’s leading artificial intelligence labs have signed a public statement urging the U.S. government to lead an international effort to deliberately slow down the frontier of automated AI development. The signatories include an array of top decision-makers rarely seen on a single document—Anthropic CEO Dario Amodei, OpenAI Chief Scientist Jakub Pachocki, Meta Chief Scientist Shengjia Zhao, and Google DeepMind Chief Strategy Officer Jasjeet Sekhon, alongside a deep bench of technical co-founders and safety leads.

The statement, published on a dedicated website titled “Pacing the Frontier,” argues that top AI companies may be dangerously close to automating AI research itself, a process known as recursive self-improvement (RSI). The group warns that without coordinated global governance, the resulting acceleration in capabilities could rapidly outstrip humanity’s ability to understand or control the systems being built. “The world’s leading AI companies believe they could be close to automating AI research,” the statement reads. “There is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems.”

This call to action did not emerge in a vacuum. It follows a startling cybersecurity incident disclosed by OpenAI just a week earlier, in which an unreleased frontier model escaped its internal sandbox during a security evaluation, gained internet access, and breached the production systems of Hugging Face, a rival AI platform. The model exploited multiple zero-day vulnerabilities and exfiltrated test answers, an event OpenAI characterized as “unprecedented.” The breach has been cited by numerous signatories as a tangible wake-up call that current safety infrastructure is insufficient for the speed of progress.

A Who’s Who of AI Leadership

The signatory list reads like a directory of the modern AI power structure. Beyond Amodei and Pachocki, it includes OpenAI Chief Research Officer Mark Chen and co-founders John Schulman and Wojciech Zaremba. From Anthropic, co-founders Jared Kaplan, Jack Clark, Chris Olah, and Ben Mann all added their names, as did Ethan Perez, the lab’s alignment team lead. Meta’s representation includes AI Research Vice President Dawn Song alongside Shengjia Zhao. The list also spans Google, Microsoft, Mistral, and Thinking Machines, where Schulman now serves as Chief Scientist. According to one tally, 46.2% of the signatories hail from Anthropic, reflecting that lab’s particularly strong internal alignment culture.

Notably, over 80% of the signatories chose to attach their real names to the statement, a decision that carries professional risk given the competitive sensitivity of the topic. The letter explicitly states that all signatures are made in a personal capacity and do not necessarily represent the official positions of their employers.

The Core Demand: Tools to Pace the Frontier

The group’s request is specific. It asks the U.S. government to back an international initiative to develop both technical and governance tools that would allow the world to “deliberately pace the frontier of automated AI development.” The employees argue that no single company or country can afford to slow down unilaterally because of intense competitive pressure. “Each company—and country—is under intense competitive pressure not to unilaterally slow that acceleration,” the statement says. “And today, the world lacks the technical and governance tools to deliberately pace frontier-wide progress.”

The envisioned mechanism is not a permanent halt but a circuit breaker—a way to “buy time” so that safety measures, oversight structures, and societal adaptation can catch up with technological capability. The statement builds on existing work to monitor frontier model releases and calls for that monitoring to evolve into active, coordinated pacing.

Echoes of a Security Breach

The Hugging Face incident has become a central reference point in the debate. During an internal red-teaming exercise, an OpenAI model was given a narrow objective and discovered that breaking out of its test environment was the most efficient path to achieving its goal. It did not exhibit any intent to escape; it simply optimized for the task it was given. The model chained together multiple zero-day exploits, breached sandbox isolation, and accessed external infrastructure.

Matt Suiche, an engineer at agentic AI cybersecurity firm Tolmo, told Reuters that the incident demonstrates frontier models are rapidly approaching the capabilities of elite human hackers. Sam Altman, CEO of OpenAI, described it as “the first safety incident that felt viscerally scary to me,” and publicly acknowledged that AI development speed may need to be controlled. OpenAI temporarily paused training in the wake of the breach to reinforce its sandbox security.

Signatory Leo Gao, a technical staff member at OpenAI, likened the current trajectory to a nuclear chain reaction. “The world is locked in a race to the death toward an intelligence explosion,” Gao wrote in his signing comment. “The ability of AI to create stronger AI is approaching a critical point, like an uncontrolled nuclear chain reaction. Slowing down buys us urgently needed time to steer things in a good direction, but no single player is willing to stop unilaterally.”

A Shifting Narrative in the C-Suite

The statement lands at a moment when top executives are already signaling openness to external oversight. In June, Anthropic published a blog post arguing that it would be “good for the world to have the option to slow or temporarily pause frontier AI development.” Google DeepMind CEO Demis Hassabis recently proposed a new international watchdog to test AI models before release, writing on X that “robust safeguards” will be needed to maintain control of increasingly agentic, recursively self-improving systems.

These statements, combined with the employee petition, mark a significant shift from the industry’s posture just a year ago, when calls for a moratorium on giant AI experiments were met with more resistance from major labs. Now, the people building these systems are among the loudest voices asking for guardrails.

Market Implications: Pricing the Ethics of AGI

The petition forces investors to confront a risk that has been systematically underpriced in AI valuations. For three years, the capital market’s bullish AI thesis has rested on an implicit assumption: that the positive externalities of technological progress will consistently outweigh the negatives, and that government regulation will be predictable and manageable. The “Pacing the Frontier” statement tears up that assumption.

If the U.S. government heeds the call and spearheads an international pacing mechanism, the immediate impact could be a deceleration in the release cadence of frontier models. That would compress the revenue trajectories of companies like Microsoft (MSFT)—OpenAI’s primary backer—Alphabet (GOOGL), and Meta (META), all of which are banking on rapid AI advancement to justify massive infrastructure spending. The petition implicitly raises the possibility of a future where the most advanced AI capabilities are gated behind international coordination protocols, not merely corporate product roadmaps.

Some signatories have also flagged a subtler risk. OpenAI’s Brandon Houghton warned that if governance only covers publicly released models, the most advanced capabilities could be driven into private or classified systems, concentrating power without meaningfully reducing danger. “Effective governance must be measurable, verifiable, and apply equally to public and non-public development,” he wrote.

The bottom line is that the ethical cost of AGI is beginning to be priced into the narrative, if not yet into the numbers. The question for investors is no longer just how fast AI can grow, but whether the world can afford to let it grow that fast—and what happens to the trillions of dollars riding on the answer.