Jon Ossoff is calling for inspectors in AI labs.
That’s necessary. But it’s not sufficient without upstream licensing for anyone using AI past a certain threshold of consequence.
An LLM is an assumption-making machine — it fills every blank with a confident guess. That’s actually how the technology works. But humans have a problem with that: when something looks credible and believable, we trust it without a second thought, even when a second thought would immediately catch the error. There’s research on this. But with AI, you can’t afford that default. You always have to trust and verify.
Do we have the political will to require accountability?
The companies building AI are failing because they do not understand humanity — or more precisely, they didn’t design for it. AI wasn’t created to be trustworthy off the shelf.
Regulation that licenses inspectors in frontier labs is necessary. But what we need alongside it is licensing for anyone using AI past a certain consequence threshold — and the credential is not coding skill. It’s the demonstrated ability to maintain and monitor the assumptions that shape behavior. Who decides what the system will guess? Who owns the fact that those guesses affect humans? How do you know they’ve thought about it?
That’s where accountability lives: not in auditing the made, but in credentialing the maker.
The question: Do we have the political will to require true accountability?