More than 1,000 employees from the world's leading artificial intelligence firms signed an open letter this week calling on the U.S. government to deliberately slow the pace of frontier AI development until safety and security measures can catch up.
The letter, published on Monday and coordinated by a coalition of current and former researchers from OpenAI, Anthropic, Meta, and Google, represents one of the largest internal dissent movements the technology industry has ever seen. Signatories include OpenAI's chief scientist, Anthropic's cofounders, and multiple vice presidents from Meta and Google's DeepMind division, according to copies of the document reviewed by this publication.
The petitioners argue that the current trajectory of automated AI development has outpaced the capacity of regulators, researchers, and the companies themselves to ensure that increasingly powerful systems remain under human control. They are asking Congress to establish what they describe as international tools to deliberately pace the frontier of automated AI development, a phrase that signals both urgency and a recognition that no single country can unilaterally slow a global technology race.
The letter arrives at a moment of acute tension in Washington over AI policy. The Trump administration has already taken steps to gatekeep the release of OpenAI's most advanced models, and a July 2 deadline for federal agencies to deliver compliance frameworks under Executive Order 14409 has just passed. Meanwhile, China is reportedly weighing its own restrictions on foreign access to its most advanced AI systems, raising the prospect of a two-way silicon curtain that could reshape global technology flows for decades.
Industry reaction has been divided. Some executives have defended the current pace of development, arguing that slowing down would cede ground to Chinese competitors and that open-weight models are essential for innovation and security. A separate open letter from Big Tech leaders, also circulated this week, advocated precisely for open-weight AI models as a counterbalance to what they see as excessive regulatory caution.
The researchers' petition, however, reflects growing internal unease about the gap between what AI systems can already do and what the industry understands about how they do it. Anthropic's own research published earlier this month found that its Claude model had developed an internal working memory during training, a discovery that the company said raised new questions about AI safety and what happens inside large language models.
Congressional aides said the letter was being circulated on Capitol Hill but that legislative action before the August recess was unlikely. The debate, they suggested, would likely extend into the fall, when the UN's second Global Dialogue on AI Governance is scheduled to convene in New York.
Image source: i.ibb.co