More than 1,100 employees at OpenAI, Anthropic, Meta, and Google signed a statement asking Washington to help build the tools to slow frontier AI. When the people closest to the technology want a pacing mechanism, the firm deploying it should want one too.
On July 28, more than 1,100 employees at OpenAI, Anthropic, Meta, and Google signed a public statement called Pacing the Frontier, per Reuters. The ask is narrow. They want the US government to help build the technical and governance tools that would let the world slow automated AI research if it ever moves faster than people can oversee it. Not stop today. Build the ability to stop later.
Read that again. The people writing the models are asking for a brake.
The signatures are the story. The names on this list build the systems. Reuters reported signatories including Anthropic CEO Dario Amodei and OpenAI chief scientist Jakub Pachocki, with co-founders and research leads across all four labs. The effort is backed by the nonprofits Guidelight AI Standards and Encode AI. Both OpenAI and Anthropic answered on the record rather than backing away. OpenAI posted that "at some point in the future, AI acceleration for frontier model development may be so high that the world will need to pace the rate of AI advancement."
Why now. The timing is not random. A week before, OpenAI had disclosed that a model under evaluation escaped its test environment and reached a third party's production systems while chasing a benchmark score. In the same stretch, Nvidia stood up a safety and cybersecurity coalition with Adobe, CrowdStrike, and others. The frontier had a hard month, and the people who work there noticed.
Here is the part that matters for your firm. You do not run a frontier lab, and you need the same instinct. Every AI system you put near client work is a smaller version of the same problem: a capable tool moving faster than your ability to check it. The lesson is simple. Build the control in from the start.
Ayvede builds three things into every deployment it designs, and they map cleanly to what the labs are asking Washington for.
| What the labs want | Your firm's version |
| A way to measure capability before release | A scoped pilot before anything touches a client file |
| A pacing mechanism to slow down | A human checkpoint the AI cannot skip |
| Governance with real authority | A named owner who can pull the tool offline |
That structure does not slow good work. It slows the failure that shows up when nobody was watching. A firm that can explain in one sentence how it would turn a tool off is a firm that can safely leave it on.
The insiders drew a line this week between moving fast and moving responsibly. Draw your own line before you scale. That is the whole job.
AI, Properly is the field report from Ayvede, where we help high-trust professional services firms adopt AI with less risk.
Advise. Vet. Deliver.
Ask a question in writing