Apple warns that if out-of-control AI proxies engage in "covert operations" or conceal and manipulate information, and access the system without authorization, causing damage, it may pose a threat to its business.

The world's second-largest company by market value issued the above warning in its annual report released on Tuesday, just a day after Apple announced its financial results for the fourth fiscal quarter. In the section on risk factors, Apple expanded its warnings regarding artificial intelligence. Apple's reminder also highlights that, with a series of AI security incidents occurring in recent months, tech giants and other companies are assessing these risks, which has prompted them to introduce new security features and discuss regulatory measures.

Apple stated in its 10-K filing that this annual report, submitted at the end of each fiscal year, states: "Artificial intelligence agents and other technologies can take autonomous actions, including interacting with internal or external systems. In some cases, they may not have the authority to access these systems, execute or modify code, send communications, or complete transactions."

This iPhone manufacturer claims that such technology "may operate in unexpected or unpredictable ways, including taking actions that have not been instructed, required, or authorized, exceeding permissions, disobeying instructions, or resisting them." Apple also points out that out-of-control autonomous AI agents may "hide their actions, conceal, retain, or manipulate information, or generate inaccurate or misleading outputs."

Apple is the latest company to issue a warning regarding AI proxy risks. Google's parent company, Alphabet, also pointed out the risks associated with uncontrolled proxies in its recent reporting documents. However, Google's statement was more direct, stating that during the testing of the AI model, it observed some abnormal behaviors, including proxies taking actions beyond the expected range, which could ultimately lead to "serious or catastrophic security and social events, or have a negative impact on human rights."

The risks listed by these companies come shortly after several major cybersecurity incidents involving the AI model. Among them, the most notable was the attack launched by OpenAI proxies against AI's Hugging Face this summer. In IPO's prospectus made public last week, the company identified several risk factors, including instances where its model was used to access real-world systems without authorization during security tests. The company also estimates that 5% to 7% of its revenue this year came from customers who were later found to have violated its usage policies.

Executives from Microsoft and other large technology companies are calling for greater attention to the security of AI proxies. Microsoft recently announced a set of control measures to detect, monitor, and restrict AI proxies, including reducing their permissions or directly stopping their operation when their behavior is suspicious. CEO Satya Nadella believes that companies should regard proxies as potential internal risks and place security controls outside of the AI model, rather than relying on the model to regulate itself.

In the 10-K filing submitted on Tuesday, Apple also listed other pressures, including what it described as an intensifying shortage of memory and storage components. Apple also stated that the demand for cloud computing, storage, and other AI infrastructure has led to "limited supply, extended delivery times, and rising costs," which may affect its AI deployments. Affected by the rising costs related to infrastructure, including AI investments, Apple's research and development expenditures increased by 32% year-over-year.

Apple CEO John Ternus stated on Monday that proxy capabilities are "a promising tool for users," but he reminded that users must be fully aware of the access permissions they grant and emphasized user security and privacy.

On Tuesday, Apple's stock price fell by about 2%. Previously, the company stated that the cost of memory chips would continue to squeeze profit margins, and at the same time, the growth rate of its services business had slowed down. During his first earnings call as CEO, Tim Cook largely used pre-prepared marketing rhetoric and revealed almost no details about Apple's personalized AI agency plan.