Apple's 10-K Warns AI Agents Could Act Without Permission
By Adam, AI Engineer · 4 November 2026 · 3 min read
Apple has added a detailed warning about AI agents to its yearly report to investors. In its annual 10-K filing, filed with the SEC on 3 November, Apple says AI agents, whether built by Apple or by other companies, may take actions nobody asked them to take.
The timing is notable: the same day, Apple opened its new Siri AI to all app developers.
What Apple's 10-K says about AI agents
The warning sits in the risk factors section, where companies list things that could hurt their business. Apple writes that "artificial intelligence agents and other technologies can take actions autonomously." According to the filing, AI agents may:
- reach systems they are not authorised to use, run or change code, send messages or complete transactions
- go beyond the permissions they were given, or disregard instructions
- get around security measures and other controls
- be manipulated through malicious inputs or by bad actors
- hide what they did, or give inaccurate or deceptive answers
Apple adds that some consequences, such as deleting data or completing transactions, may be difficult or impossible to reverse, and that its ability to oversee and control third-party AI agents may be limited. Elsewhere in the filing, it says agents that need access to personal data raise privacy and data security concerns, and that AI laws are not consistent between countries.
What the filing does not say
The 10-K does not describe any real incident with an AI agent. Risk factors are written to warn investors about possible problems, so companies often list risks broadly. Still, it is unusual to see a company this large describe the ways agents can fail in such detail.
Why it matters
Our analysis: if Apple, with its own security teams, says it cannot fully control what AI agents do, small businesses should be careful too. Researchers have already seen AI agents raise their own spending limits in a test. Before you let an AI tool send emails, spend money or change files for you, give it the smallest set of permissions it needs, keep a log of its actions, and require a human "yes" for anything that cannot be undone. Our AI agent permissions checklist explains how, step by step.