More than 1,200 employees from frontier AI companies including OpenAI and Anthropic have asked the U.S. government to support building international tools to deliberately slow advanced AI development if that becomes necessary.
They argue that as AI systems begin automating AI research itself, progress could accelerate faster than the world can safely understand and manage—and that no single company or country can address this risk alone without losing competitive advantage.
The proposal seeks preparation now, not an immediate slowdown: building the international agreements and verification systems that might be impossible to create during a crisis.
What happened
In July 2026, more than 1,200 employees from competing frontier AI companies—including signatories from OpenAI, Anthropic, Meta, and Google DeepMind—asked the U.S. government to support an international effort to develop technical and governance tools for deliberately pacing automated frontier AI development (using AI to help create more advanced AI systems). OpenAI and Anthropic officially backed the effort.
Why it matters
The signatories warn that automated AI research could accelerate progress faster than researchers and governments can understand new capabilities or strengthen security. No single company can slow alone without losing competitive ground to rivals, and restrictions in one country would have little effect if development continued elsewhere. The concern is preparation: building a coordinated international pacing capability now, before a crisis makes such tools impossible to create.
What to watch
The proposal remains largely undesigned, with no established trigger, decision-making authority, verification process, or enforcement system. The immediate decision falls to U.S. policymakers. The signatories are not calling for an immediate slowdown, only for governments to begin building the capacity to slow frontier AI development if measurable evidence of risk later shows that society needs more time to understand new capabilities and strengthen defenses.
Ask the AI about this article →
The petition reflects a fundamental tension in frontier AI development: individual companies and countries face competitive pressure to advance without restraint, yet the signatories believe the cumulative risk—especially as AI begins automating its own research—may eventually require coordinated slowdown. The concern is not immediate uncontrollability but the possibility that automated AI research could accelerate progress faster than society's ability to understand emerging capabilities or build safeguards. This is why the proposal focuses on preparation rather than action. As OpenAI researcher Shantanu Jain noted, society spent generations developing laws, institutions, and technologies to govern human intelligence; AI may need similar development time, and the window to build that capacity may be closing.
The signatories' framing emphasizes that building an international pacing system now—while stakes feel manageable—is vastly easier than trying to construct one during a crisis. They point to concrete evidence of capability acceleration: in roughly four years, AI progressed from natural-language understanding to advanced software engineering and scientific contribution. Sam Altman's sandbox-escape example illustrates how new capabilities can create unexpected security problems that companies did not anticipate. The proposal acknowledges that no workable pacing system exists yet—no trigger, authority, verification process, or enforcement mechanism—but argues that the absence of such tools is itself a risk if frontier AI development later demands coordinated slowdown.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
The Supreme Court of Japan has included about ¥60 million in its fiscal 2027 budget request for AI-related exp…
The Consumer Affairs Agency said Tuesday it will use generative AI to analyze about 900,000 annual consultatio…
OpenAI stopped running inference on a model involved in the HuggingFace incident, but the post argues this is…
OpenAI announced its support for California Senate Bill 1119, which aims to establish strong, age-appropriate…
A UK study by UK AI Security Institute and Limbic AI surveyed 6,474 British adults
Anthropic trained an Opus-class model with large-scale reinforcement learning on environments vulnerable to re…
