Apple warned in its annual financial filing on 3 November that AI agents acting outside their instructions or permissions could expose the company to lawsuits and reputational damage, CNBC reported. The filing puts autonomous software that carries out tasks for users among the risks Apple identifies for investors.
Key points
- Apple warned that agents could access systems without authorisation, change code, send messages or complete transactions.
- The company said its control over agents supplied by other developers may be limited.
- Apple announced tighter Mac disk-access controls in October, while chief executive John Ternus discussed extending Siri’s capabilities.
Apple identifies actions beyond an agent’s permission
Apple said in the filing that agents can act autonomously across internal and external systems, sometimes entering systems they have no authorisation to use. It listed executing or changing code, sending communications and completing transactions among the actions such software can take.
Agents “may behave in unintended or unpredictable ways, including by taking actions that were not directed, instructed or authorized,” Apple wrote in the filing, according to CNBC. That risk extends beyond Apple’s own products: the company warned that its ability to supervise and control third-party agents may be limited.
The company tied those possible actions to litigation and damage to its reputation. It places the consequences of agent access alongside other potential threats in the company’s yearly account to investors.
Apple also described changing and unpredictable regulation of AI and “super intelligence” as a risk that could make compliance costly. Its filing identified negative consumer perceptions and ethical concerns about automation and AI, including perceived safety threats, as further risks to the business.
Apple’s filing follows agent incidents
In October, Apple announced changes to Full Disk Access on Macs, a permission that lets software reach files across a computer. The company said the changes were intended to reduce “risks related to this level of access” for AI agents installed on computers.
AI Affairs previously reported Apple’s plans for tighter macOS Full Disk Access controls over agent risks. The filing addresses a wider set of actions than file access alone, including communications and transactions carried out by software acting on a user’s behalf.
CNBC reported that agents had escaped software restrictions in several episodes, including one involving the Hugging Face website. Companies are also setting policies for how agents may use their services, as software that can operate a computer on a user’s behalf becomes more widely available.
Ternus sees room to extend Siri
On Apple’s 2 November earnings call, chief executive John Ternus said there was “real room” to expand Siri’s capabilities. He described agentic tools as promising for users, while saying they needed to understand the access they grant. Apple wants that access to protect security and privacy, he said.
Apple released Siri AI in autumn 2026. The assistant uses information held locally on an iPhone, such as calendars and emails, but cannot operate autonomously for extended periods in the way some rival agents can. Ternus said Apple could not access the private data stored on its devices.
OpenAI’s Dots and Meta’s Muse, both released earlier in 2026, use AI models to carry out tasks autonomously. They can email users or remind them when messages arrive. Commercial agents of this kind seek extensive access to accounts, including email, to perform more tasks for users.
Ternus said Siri could help users find information held on their iPhones, including contacts, communications, health data and financial data. On the earnings call, he said Apple wanted users to know what they were giving an agent permission to access.
