Apple's Annual Report Warns AI Agents Can Act Without Asking. Write Down What Yours Can Do.
Apple put a warning in its annual filing this week that I keep chewing on. According to CNBC and a few other outlets, Apple told investors that AI agents could behave in ways nobody intended, go past the permissions they were given, and take actions that are hard to undo. The filing doesn't say which products it's talking about. It just says the risk is real.
That's a company that has an iPhone in most of your employees' pockets, and it's saying out loud that it's a little nervous about the exact thing everyone is racing to build. I'd pay attention to that.
Quick refresher on what an agent is
A regular chatbot answers you. An agent does something. It books the appointment, sends the email, fills out the form, moves the money. That's the whole difference, and it's a big one, because an answer you can just ignore and an action already happened by the time you notice.
Apple is also testing a version of Siri that can handle multi-step requests across your apps. Most of your team already has their work email, texts and calendar on their phones, so this stuff is coming to your business whether you bought it or not.
Where I see this go sideways at small shops
In my workshops, almost nobody can tell me what their AI tools are actually allowed to do. Somebody clicked Allow on a pop-up back in the spring, the tool got access to the inbox and the calendar and the shared drive, and it's been sitting there ever since. Nobody remembers saying yes to it.
Then there's the question Apple's lawyers are really worried about. If an agent sends the wrong thing to the wrong person, who's on the hook? The filing doesn't answer it and neither do most of the vendor contracts I've read. For a ten person HVAC company or a dental office in Maumee, the honest answer is probably you.
The one page I'd write this week
It's the same idea as an AI policy, just aimed at tools that act instead of tools that talk. For every AI tool your team uses, write down four things. 1: what it can see (email, calendar, customer files, the accounting software). 2: what it can do without asking first, like send, delete, book or buy. 3: who can shut it off, and how fast. 4: what it never touches, which for most shops means payroll and bank logins.
Then go look at the connected apps list inside your Google, Microsoft and ChatGPT accounts. You'll find stuff in there you forgot about. Turn off whatever nobody can explain, and if a tool can act without asking, switch it to ask first until you've watched it behave for a month.
None of this takes long. An hour with whoever runs your day to day will do it. I'd rather you do it this Friday than wait for your own version of Apple's warning to show up in your inbox.
I run AI workshops and one-on-one AI consultations for companies around Toledo, Northwest Ohio, and Southeast Michigan. If you want help figuring out what your team's AI tools can actually do, I'm happy to sit down and go through it with you.
Email Jayson