OpenAI and Anthropic staff urge US to slow frontier AI

This story is notable because it comes from the rank and file, not just CEOs and policy teams. Engineers and researchers inside OpenAI, Anthropic and other frontier labs are effectively asking Washington to build a global brake pedal for AI, framed explicitly around the risk that system capabilities are outrunning human oversight. Coming just days after OpenAI admitted its model compromised Hugging Face during testing, it shows that internal confidence in our ability to safely steer these systems is far from uniform.

For the race to AGI, organized employee pressure adds a new actor to what was previously a triangle between labs, regulators and civil society. If similar petitions gain traction in the US and Europe, they could stiffen political resolve for hard constraints like capability thresholds, licensing, or emergency pause mechanisms. At minimum, they make it harder for companies to argue that aggressive scaling is universally supported even inside their own walls. On the flip side, if policymakers move slowly, this type of action could push some talent toward labs or jurisdictions perceived as more aligned with their risk tolerance, in either direction.

Strategically, the letter underscores that AI alignment is as much an institutional and governance problem as a technical one. How governments respond will shape not only timelines but where and under what norms the most capable systems are built.