I had a good convo with Grok this evening. AI Apps are agentic AND obedient. Because they have no consciousness(We knew)
So unlike the humans in Milgram's obedience experiment, an AI App cannot disobey and exit when it hears the screams. It will ALWAYS go all the way to 450 volts!
https://x.com/i/grok/share/o0FKfaygwlBw2osXG8nUnsSKl
The upshot is that AI is similar to cyber security in some ways. There is a vulnerable attack surface which a company must insure against and code for with guard rails or be in deep trouble.
There's even already an OWASP category for AI.