When you need “A.I.” to keep other (or your own) “A.I.” in check, you’re not in control of your technology anymore. Three steps from the starting line.
And the arms race (both figuratively and literally) makes it impossible to think twice. Nobody is going to give up a stupid idea if a competitor or adversary has a more powerful stupid idea.
There’s a difference between “unable to control technology” and “criminally negligent failure to follow industry-standard security practices.”
Human sloppiness certainly doesn’t help, that’s for sure.
The problem is that “agentic ‘A.I.’” will get access to all kinds of stuff (most of it harmless by itself), not by accident or plotting, but intentionally granted. Otherwise it’s of no use.
I’d call the problem-solving strategies already exhibited by this still relatively dumb generation of “A.I.”, in human terms, psychopathic behavior. (“A.P.”? 
)
Which is exactly why agentic AI is going to be a tough sell without much better guardrails than are currently in place.
They should add a moral compass, that would be a nice feature. 
This has been one of the most baffling parts of the current headlong rush: the number of people who seem happy to give control of their devices and data to a black box algorithm with a high but unpredictable error rate and a penchant for ignoring instructions, even if it does apologise nicely afterwards.
Particularly when most of the touted benefits appear to be that it lets you do things I wouldn’t want or need to do, or would prefer to do for myself. That’s possibly because I’m not the target audience, but still, you’d expect some sense of basic self-preservation to kick in at some point, wouldn’t you?
People are looking for a “personal assistant” without having to pay a human. In particular, they don’t want to pay a human for instant action on whatever whim pops into their head 24/7.
Never mind that “sanity check” is an important part of a good human assistant’s job.
You guys know that the US military almost started a war with China, right.
Just this last spring.
AI randomly said a specific chinese ship carried nukes or something, and that information was used as intel.
