Apple's annual report adds new warnings that AI agents could act on their own, hide actions or defy instructions
The Cupertino company's yearly filing lists AI agents as a business risk for the first time.
Apple warned investors this week that AI agents could behave in ways it cannot fully control, in its yearly report to federal regulators.
The report, called a 10-K, is a detailed annual filing public companies must send to the Securities and Exchange Commission. Its "risk factors" section lists things that could hurt the business. Apple filed it Tuesday.
AI agents are programs that do tasks for a person instead of just answering questions, such as booking a trip or sending an email. Apple wrote that agents "can take actions autonomously," including reaching systems they may not be allowed to access, changing code, sending messages and completing purchases.
Apple said agents may act unpredictably, including "exceeding permissions, defying directions or resisting instructions" and getting around security measures. It said they could "conceal actions, hide, withhold or manipulate information," can be tricked by malicious inputs and may take actions that cannot be reversed. Because the same request can produce different results each time, small errors can build on each other, Apple wrote. It also said it has limited oversight of agents made by other companies.
Apple said agents' access to personal data "creates privacy and data security concerns that could erode customer trust." Apple's 10-K for the previous year did not use the word "agent."
Source: Apple Form 10-K for fiscal 2026, filed Nov. 3, 2026 (SEC)