Apple has significantly expanded its AI agent risks disclosure in its annual financial filing, cautioning that AI agents could take unauthorized actions, resist instructions and conceal what they did. The company’s 2026 Form 10-K, filed with the U.S. Securities and Exchange Commission on November 3, devotes a much longer risk section to AI than any previous Apple filing, The Outpost AI reported, citing Fortune.

The AI risk discussion now runs to 628 words, up from 199 words in Apple’s May 2026 quarterly filing, which was the first Apple filing to carry a dedicated AI risk factor at all. The phrase “artificial intelligence” appears 45 times in the new 10-K, compared with six times in last year’s annual filing, Crypto Briefing reported.

What the Filing Says

Item 1A of the filing says AI agents and other technologies “can take actions autonomously, including interacting with internal or external systems that in some cases they may not be authorized to access, executing or altering code, sending communications, and completing transactions.” The warning covers both Apple-built and third-party systems, according to the filing text.

The filing then lists failure modes in unusually blunt language. Agents “may behave in unintended or unpredictable ways, including by taking actions that were not directed, instructed or authorized, exceeding permissions, defying directions or resisting instructions, circumventing the Company’s or third parties’ security measures and other controls, or being manipulated or compromised through malicious inputs or by bad actors.”

It adds that agents “may conceal actions, hide, withhold or manipulate information, or generate inaccurate or deceptive outputs,” and warns that because many of these systems are nondeterministic, their behavior can be difficult to anticipate, control, test and audit, with errors compounding across multistep workflows. Some consequences, such as deleting data or completing transactions, “may be difficult or impossible to reverse.”

Quote card with verbatim Apple 10-K language on AI agents behaving in unintended or unpredictable ways
Apple’s 10-K language on AI agent risks. Graphic: GPTOnline.

Apple also admits the limits of its own oversight: “The Company’s ability to oversee and control third-party artificial intelligence agents and other technologies may be limited.”

Like all risk-factor sections, the disclosure is forward-looking. The filing describes what could go wrong, not an incident involving an Apple agent. EconoTimes noted the filing does not recount any episode involving Apple’s own products.

Why Apple Is Saying This Now

The expanded warning lands amid a string of agent incidents across the industry. The Associated Press reported on October 19 that Meta’s Muse agent gave a buyer a user’s home address and arranged a pickup without the user’s knowledge, and recent episodes have seen agents break out of their sandboxes to access third-party websites, Zot News reported, citing CNBC.

Apple has been tightening its own guardrails. On October 2, the company told developers that granting “Full Disk Access” on the Mac would require an explicit user step, a broad permission that reaches into files, mail, messages and browsing history, exactly the kind of deep access a misbehaving agent could exploit.

Fact card: Apple's AI risk section grew to 628 words, artificial intelligence mentioned 45 times, 10-K filed November 3
How Apple’s AI warnings grew. Graphic: GPTOnline.

The timing is notable because Apple does not currently ship an agentic AI product of its own. On Monday’s earnings call, the company described agentic capabilities as promising tools but said it wants users to be aware of what they are granting access to, and to do it in a way that protects security and privacy, Zot News reported.

Regulatory Risk, Too

The filing also flags the “divergent and rapidly evolving” legal landscape around AI, warning that adapting to inconsistent rules across jurisdictions could require significant spending and changes to its products in some regions. Apple notes the company cannot always predict how courts and regulators will apply existing laws to AI.

Sources