Why is Apple warning about rogue AI agents?

It does not even build the biggest AI models. So why is its annual report worried about agents that ignore instructions?

Apple's annual report now says, in plain words, that AI agents can behave in unintended or unpredictable ways. They may take actions nobody asked for, go beyond their permissions, or ignore instructions. Some of the damage, like deleted data or completed transactions, may be hard or impossible to undo.

Fine, but Apple is hardly the one building these things. Why does it care?

Because it is not just a bystander. Its voice assistant now runs on a model from another company, and Apple is opening that assistant to developers so it can perform actions inside other people's apps. That makes Apple a deployer, not only a user.

Then there is the bit that explains the tone. The warning covers agents made by Apple and by third parties alike, and it admits that its ability to oversee and control other people's agents may be limited. Apple owns the operating system that everyone's agents will run on. If one of them goes off the rails on an iPhone or a Mac, it will not feel like somebody else's problem to the person holding it.

One more thing worth knowing: risk sections in these reports are mandatory lists of what could hurt the business, not mission statements. But this one reads like a company that has looked at where all of this is heading and decided to say so first.