> This seems the exact opposite of what the current crop of AI is good at.

Yeah, last thing I want when trying to share a file with people outside of my organization is some opaque kafkaesque chatgpt-wannabe model telling me "I’m sorry Dave, I’m afraid I can’t do that".

"I'm 99% certain that you're a poodle, and poodles are not allowed to share that document. Is there anything else I can help you with today?"

The AI doesn't need to be the end-all-be-all of defining permissions. But it's not hard to imagine a couple of areas where AI could help:

1. Letting users enter desired permissions setting in a natural language, and then the AI recommending checkbox settings, and, importantly, explaining these settings.

2. Useful as a monitoring/alerting system for DLP. Most DLP systems already use some sort of machine learning for identifying sensitive docs.

3. Easily running "test scenarios" to show to admins who can and can't get access to particular docs.

There is a huge chasm between "AI owns all my permission settings" to "AI can make it easier and more robust for me to understand what my permission settings should be."

There is a reason permissions are not defined in natural language. Language is imprecise and that is why security is so hard to get right