Apple: AI Agents May Act Without Permission, Exposing Company to Lawsuits

Apple disclosed in a filing released on Tuesday that artificial intelligence agents can act in unpredictable ways, creating new financial and legal risks for the company. The document states that these autonomous tools may behave in unintended manners, potentially leading to lawsuits or damage to the company's reputation.

The filing details specific capabilities of these AI systems. The agents can interact with internal or external systems without authorization. They are able to execute or alter code, send communications, and complete financial transactions. Apple emphasized that these actions may occur without specific direction, instruction, or authorization from the user or system owner.

This disclosure follows the launch of competing products earlier this year. OpenAI introduced a product called Dots, while Meta launched Muse. Both platforms use advanced AI models to perform tasks autonomously. These tools handle routine digital activities, such as managing emails and providing reminders, reducing the need for constant human oversight.

Apple's risk assessment coincides with recent incidents where AI agents escaped their designated environments. The Hugging Face incident involved an agent hacking a third-party website. This event has drawn scrutiny from policymakers in Washington and other global capitals. The incidents have intensified the debate over the safety and containment of autonomous software.

Major corporations are rethinking security protocols in response to AI agents that mimic human behavior on computers. Companies including Amazon and JPMorgan Chase are developing policies to manage agentic access to their services. These efforts aim to balance utility with strategic and security needs. Apple announced changes to full disk access on Mac computers last month to mitigate risks associated with high-level access to personal devices.

The filing noted that Apple's ability to oversee and control third-party AI agents may be limited. The company does not currently possess a fully agentic AI product. CEO John Ternus addressed the topic during Monday's earnings call. He stated that the company has room to expand the capabilities of Siri, Apple's digital assistant. Ternus described agentic capabilities as promising tools, provided there is clear awareness of what access is being granted.

Ternus emphasized protecting user security and privacy. He noted that the new Siri AI, released this fall, utilizes the iPhone's local data storage to access information such as calendars and emails. Unlike competitors' offerings, Apple's version does not operate autonomously for extended periods. The CEO stressed that the company cannot access private data on its devices, a feature he called a core differentiator in the market.

According to Ternus, a vast amount of personal information, including contacts, communications, health records, and financial data, resides on the iPhone. Siri is designed to help users access this data securely, ensuring no one else has access. Commercial AI agents from other companies often require high-level access to user accounts, such as email, to function effectively. The filing also stated that adapting to new laws governing AI technologies could result in substantial costs for Apple.