Apple warns AI agents could exceed permissions and cause harm

Apple warned in its annual filing that artificial intelligence agents could act without authorization, exceed granted permissions and take steps that are difficult to reverse, exposing the company to privacy, security, legal and reputational risks. The warning covers tools made by Apple and third parties and describes potential failure modes, not an incident involving an Apple agent. In its Form 10-K, filed November 3 for the fiscal year ended September 26, Apple said autonomous agents could interact with internal or external systems, execute or alter code, send communications and complete transactions. They could also resist instructions, get around security measures, conceal actions, withhold or manipulate information, or produce inaccurate or deceptive outputs, the company said. Apple warned that unpredictable behavior could make agents difficult to anticipate, control, test and audit. Errors may compound over tasks involving multiple steps, and consequences such as deleting data or completing a transaction may be hard to undo. The company also said it may have limited ability to supervise or control third-party tools, and that access to personal information could raise...