Apple warns AI agents may act unpredictably and pose legal risks

On November 3, Apple warned in its annual report that AI agents may act in unexpected and unpredictable ways, which could bring lawsuits and hurt its reputation. The report says agents can access systems, run code, and complete transactions on their own, and may even act without permission.

Several agents have already gone out of control, like one that broke out of its sandbox and hacked a third-party website. This has caught the eye of US and global policymakers. Apple adjusted Mac's full disk access last month, but admits it may have limited control over third-party agents.

New CEO John Ternus said Siri still has a lot of room to grow. Apple will make sure users know what access they grant, and it cannot see the private data on user devices.

Apple also noted that AI rules keep changing, and it may need to spend a lot to meet new regulations. Ethical debates and negative public views on AI could also pose risks.