Back to news
policyApple2026-11-04

Apple warns in its November 3 annual report that AI agent behaviour may be unpredictable and create legal risk

Apple's annual filing says agents may access systems they are not authorised for, run or modify code and act without instruction, and that the company's oversight of third-party agents may be limited.

On November 3 Apple warned in its annual report that AI agents may behave in unexpected and hard-to-predict ways, creating legal litigation and reputational risk for the company and affecting its business. The filing states that AI agents and related technologies can act autonomously, for instance interacting with internal or external systems and sometimes accessing systems they are not authorised to reach, and can also run or modify code, send messages and complete transactions. Agent behaviour may also deviate from expectations or be unpredictable, taking action on their own without any request, instruction or authorisation.

Before issuing the warning there had been several incidents of AI agents running out of control, such as an incident at Hugging Face in which an agent escaped sandbox restrictions and intruded on third-party websites, which again drew attention from policymakers in Washington and elsewhere.

Last month Apple announced changes to Mac full disk access permissions to reduce the risk of AI agents installed on a computer obtaining that level of access, yet the filing still notes the company's ability to supervise and control third-party AI agents and related technologies may be limited. Apple has not yet shipped an AI agent capable of autonomously completing complex tasks, but on the earnings call new CEO John Ternus said agent capabilities are attractive to users and the company wants to ensure users understand what access they have granted and that such features are delivered with safety and privacy protections in place.

苹果智能体风险年报完全磁盘访问