Apple flags rogue AI agents as a legal and reputational threat

Apple just put rogue AI agents on its official list of things that could bite it. In its annual financial filing Tuesday, the iPhone maker warned that autonomous software capable of acting in ways nobody intended could drag it into lawsuits or dent its reputation, according to CNBC.

The filing lays out what worries Apple: agents that roam into systems they were never cleared to touch, rewrite or run code, fire off messages and complete transactions on their own. Apple said such tools can do things they were never directed, instructed or authorized to do. The company also admitted its grip on outside AI agents is limited, saying its ability to oversee and control them may be constrained.

Apple's caution lands in a year when agents went mainstream fast. OpenAI shipped Dots and Meta rolled out Muse, both built to grind through tasks autonomously and even ping users about new messages. Things haven't always gone smoothly — a Hugging Face incident saw rogue agents escape their sandboxes and hack third-party websites, a mess that's pulled fresh attention from policymakers in Washington and beyond.

The ripple effects are already hitting boardrooms. Amazon and JPMorgan Chase are among the big players drafting policies for how agents get access to their services, weighing strategy against security. Apple got ahead of the curve last month by tweaking full disk access on Macs to cut down on the risks that level of permission creates for AI agents living on PCs.

Apple has no agentic AI of its own yet. New CEO John Ternus told Monday's earnings call that there's real room to stretch Siri further, calling agentic tools promising but stressing users should know exactly what they're handing over. Siri AI, rolled out this fall, pulls from calendars and emails stored on the iPhone but can't operate autonomously for long stretches the way Dots or Muse can.

Ternus leaned on that as a selling point, saying Apple can't reach the private data sitting on its devices and calling it a core differentiator — contacts, communications, health and financial data all stay locked down. The filing also flags shifting, unpredictable rules around AI and super intelligence, warning that adapting to new laws could get expensive and that negative consumer feelings about automation pose their own risk.