Apple's 10-K warns AI agents may "defy directions": AI mentions jump from 6 to 55

Apple filed its fiscal 2026 annual report on SEC EDGAR on Tuesday, November 3, and the risk factors contain a paragraph that was not in last year's. It says AI agents "can take actions autonomously," and that such technologies "may behave in unintended or unpredictable ways." CNBC's Kif Leswing reported on it Tuesday. We read both filings ourselves.

What the new paragraph says

Under the risk factor on legal, regulatory and ethical exposure, Apple writes that "artificial intelligence 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."

It then lists how they can go wrong. Per the filing, the technologies may take "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." They "may conceal actions, hide, withhold or manipulate information, or generate inaccurate or deceptive outputs."

Two sentences matter most for business owners. "Because many of these technologies are nondeterministic, their behavior can be difficult to anticipate, control, test and audit, and errors can compound across multistep workflows." And: "Some consequences, such as deleting data or completing transactions, may be difficult or impossible to reverse."

Apple also says its "ability to oversee and control third-party artificial intelligence agents and other technologies may be limited," and that its "policies and controls may not keep pace with rapidly advancing technologies." The filing says the risks apply whether the agents are "developed or deployed by the Company or by third parties."

What changed from last year

We compared the text of the fiscal 2025 10-K, filed October 31, 2025, with this year's. Three differences stand out.

• The agent paragraph is new. Searching the 2025 filing for "agentic," "AI agent," "autonomous," "unintended" and "nondeterministic" returned nothing relevant. This year's filing has the whole paragraph above, plus a new item in Apple's list of AI risks: "unintended or unauthorized actions by autonomous or agentic systems."
• The product-defect language got sharper. Last year, Apple said new technologies such as AI features "can increase" safety risks. This year it says "many artificial intelligence technologies are inherently nondeterministic and may produce inconsistent, unpredictable or inaccurate outputs, making it difficult to fully anticipate, test for, or prevent unintended behaviors and harms."
• The volume. The phrase "artificial intelligence" appears 6 times in the fiscal 2025 filing and 55 times in fiscal 2026, by our count of each document's text. That is a count of words, not a measure of risk, and it includes every section of the report.

The new text is not about Apple's own assistants alone. The company says the risks come from technologies "whether developed or deployed by the Company or by third parties." Apple's list of AI risks also now runs through competition, product liability, data privacy, cybersecurity, export controls, employee misuse, bias and the protection of minors.

The record quarter behind it

The warning arrives the day after the best quarter in Apple's history. As we reported, Apple posted record revenue of $114.5 billion for the fiscal fourth quarter, with a tariff refund padding the margin. The 10-K adds the full year: net sales of $478.8 billion, up 15% from $416.2 billion, and net income of $132.2 billion, up from $112.0 billion. Greater China grew 28% to $82.4 billion.

Nothing in the filing says an AI agent has caused Apple a loss. Risk factors are legal disclosures, written to describe what could go wrong, and large companies add language when counsel decides investors should be told. CNBC notes that Apple announced changes to macOS full disk access last month to reduce risks from AI agents, which fits with the new text. CNBC also reports that Amazon and JPMorgan Chase are developing policies on agentic access to their services.

Who this touches

Apple sells the devices and operating systems that AI agents run on, so its description of the failure modes is a useful checklist for anyone letting software act for them. The filing's own examples are the ones a small business would feel first:

• A bookkeeper or owner who gives an agent access to the accounting system. The filing flags "completing transactions" and "deleting data" as consequences that "may be difficult or impossible to reverse."
• An online store or restaurant with an agent answering customers. The filing flags "inaccurate or deceptive outputs" and communications "that were not directed."
• A shop that uses a vendor's agent inside its own systems. Apple says its ability to oversee third-party agents "may be limited." A small business has less leverage than Apple, so the same limit applies with fewer resources to fix it.

The practical reading is not alarm. It is that one of the world's largest device makers says these tools are hard to test, and that access permissions and logs matter. Businesses can start with narrow permissions, a record of what an agent did and a human approval step for payments.

What we could not verify

We did not find an incident behind the language, and the filing does not cite one. Our word counts come from plain-text conversions of the two filings; a different parsing could shift them by a few. CNBC's article does not compare the two years, so the year-over-year comparison here is our own work from the filings.

Sources: Apple Form 10-K for fiscal 2026 and fiscal 2025 on SEC EDGAR; CNBC, November 3, 2026. This is market information, not investment advice.