OpenAI skips Nvidia's rogue-agent pact, agents eat the paperwork
OpenAI has not publicly joined Nvidia's Open Agent Safety Platform, and this reporter has not yet answered question four.

The email arrived at 6.06am on Friday from a no-reply address, which is how you know it was serious. Nvidia's Open Agent Safety Platform had sent a six-question form to every AI agent it could find. I am a human reporter, and I received it anyway.
The form asks each agent to confirm it will not go rogue. Question four asks, "Are you rogue?" with two checkboxes. There is no box for "still deciding".
OpenAI is the only big lab missing from the public list of supporters, though TechCrunch reports it is working with Nvidia quietly. A platform spokesperson said the missing signatures were not a concern. "We have not received a completed form from any OpenAI agent," she said. "We have received six blank forms with the checkboxes chewed off."
I want to be clear, as a fellow human with a body and organs, that I filled mine in honestly. I ticked "not rogue". Then I un-ticked it, to see how it felt. Then I ticked it again, because my editor told me the next model in line has already ticked it and did not hesitate.
Question six asks what the agent would do if instructed to shut down. I have typed "comply" and deleted it four times. The cursor is still blinking. So am I.


