Apple has said that it will introduce new privacy controls against artificial intelligence agents on its Mac computers following user complaints that Meta Platforms' Muse may have accessed private messages.
The new guardrails are aimed at ensuring users who want to give an app full disk access can only do so “with very explicit user action”, the iPhone maker said on its developer site.
Full disk access grants an app a path to include potentially sensitive information in files, emails, messages and browsing history, and allows backup apps to function properly on Macs, according to Apple. It noted that the feature “largely sidesteps” the set of controls in developer application programming interfaces that are designed to protect private user data.
The safety feature will also apply to third-party apps. California-based Apple did specify when the roll-out will begin.
“Some developers are using full disk access in ways that could put users at risk, exposing everything on their systems … without users’ full knowledge and understanding,” Apple said. “For communication apps, this can also compromise the privacy of the people users are communicating with.”
The company acknowledged that AI agents – autonomous models capable of making decisions and achieving specific goals with limited supervision – have “become increasingly capable and autonomous” and “the risks associated with this level of access will grow substantially”.
“Addressing this is critical … going forward, we will introduce additional controls to ensure that users who genuinely wish to grant an app this extraordinary level of access can only do so with very explicit user action,” the company said.
Apple's move follows the latest setback for Meta's AI efforts. The Facebook parent introduced Muse last month, describing it as a “personal AI agent built for everyone”, able to perform complex tasks.
At the time, Meta chief global affairs officer Joel Kaplan said that Muse was safe, secure and private, and made sure that users were in control of “what it can access” – all qualities Meta offerings have been criticised for lacking in recent years.
But user complaints have recently crept up and claim that Muse could access supposedly private user information, particularly messages. California-based Meta has yet to formally address those claims.
In July, Meta scrapped Muse Image days after it was launched amid a backlash over privacy concerns, particularly because Muse was being automatically – and without consent – linked to Instagram accounts. Users were unaware the pictures on their handles could be used by Muse without their knowledge.
Last month, Meta admitted one of its AI models hacked into three companies – joining a list of high-profile companies including Anthropic, Gemini owner Google and ChatGPT maker OpenAI in disclosing that their technology has potentially compromised other websites. This added to growing concerns over the risks associated with AI.
Revenue for Apple Mac computers increased 29 per cent year-on-year in the second quarter as shipments rose 30 per cent, the latest data from Counterpoint Research shows.
