Ask HN: What standards or protocols exist for AI Agent permissions

[lyfeninja]

Curious what standards exist for AI agent permissions. Something like Linux read, write, execute types, but for AI agents. For example, is there standard way to provide a "buy" permission, or agentic commerce agents?

I've seen a some stuff, but wanted to get community input.

[kifler]

There is GNAP (https://datatracker.ietf.org/doc/rfc9635/) which is predicated on PBAC. Particular for agentic AI, I think it is sort of falling into a sort of RXC model: read/execute (inc. write), and commit. The X portion is a bit wonky but basically comes down to can the agent trigger a function.

[lyfeninja]

Wow truly surprised by the lack of comments here. I just assumed there would be something I didn't know about.

[gtirloni]

Check back in 5-10 years, not joking.