bad AI hair day
Responsible businesses are absolutely right to look askance at any AI features an app in their portfolio has added. The way businesses "look askance" at things is to add pieces to their internal compliance, security, and privacy loops that introspect whatever the hell a given vendor has come up with.
Because everybody wants to be able to say they're doing something with AI, we're getting a mix of things that are genuinely hair-raising -- some formerly contained and modest application now wants to start hoovering your data into an LLM -- and things that are mind-numbingly trivial and generally sandboxed -- glorified predictive typing.
It's in the nature of SaaS for the application itself to be part of the vendor's marketing operation, so the features are just enabled "for you," and something as routine as buying a few extra seats of something mid-term triggers the internal "looking askance" part, and you're left digging through vendor docs to figure out what this "AI feature" even does so you can satisfy the privacy and security folks, who are right to ask, because the people imposing these features are children who don't know much but do know there is now a market imperative to appear to "have AI."
And because it's in the nature of businesses to assume that their demands for your attention and time should be your highest priority, they really don't care what internal IT, privacy, or security teams are left dealing with when the vendor rolls out a new thing and markets it straight into your user community.
Meanwhile, Bob over there has decided he can't live without AI glorified spell-check, so "you can just turn it off" means the enterprising vendor has signed you up for yet another conversation with Bob when his toy goes away.
I laid down a principle when I led a content marketing team that we were to give something to the reader/customer before we asked for something. Meaning, yes, we needed to move 'em down the funnel, get them to do a trial download, cough up an address for an ebook, whatever; but before we could get to the call to action, we had to give them something: A way to solve a problem, information they could use that didn't necessarily benefit us, etc. I said we would never, ever confuse what we needed with what the customer needed but would instead understand the ways in which those sets of concerns were discrete and only sometimes aligned.
That's a principle borne out of consideration for other human beings and their time and attention, informed by a confidence that you're selling useful things. So it makes sense that most companies piling these "features" into their marketing platf... their product aren't interested in the impact. Most of them are selling bunk and robot parlor tricks, so no need to be precious about it, I guess.