Meta is preparing to bring its Private Processing technology to its AI-powered glasses, allowing demanding artificial intelligence tasks to be handled in the cloud without giving Meta direct access to the underlying user data.
The company announced the planned expansion at Meta Connect on 23 September, as concerns continue over privacy and the use of cameras and AI assistants in wearable devices. The technology is designed to let AI glasses perform more advanced tasks while keeping the information needed to complete them inside a protected computing environment.
Meta originally developed Private Processing for its AI services on WhatsApp and the Meta AI app. Meta says the system can process sensitive AI requests in an environment that even the company itself cannot access. Its expansion to AI glasses could let more personal and computationally demanding features run without exposing the user’s information to Meta.
Private Processing moves demanding AI tasks into a protected environment
Many basic functions on AI glasses can already be handled directly by the device. Calls, messages and other relatively simple operations do not necessarily require large amounts of cloud computing. More advanced functions, however, can require substantially greater processing power, particularly when the glasses are expected to understand conversations, maintain context or work with more capable AI models.
Meta’s Private Processing architecture is intended to move these workloads into confidential virtual machines, known as CVMs. These machines operate inside Trusted Execution Environments designed to isolate the processing from the rest of the infrastructure. The approach uses hardware-based protection to prevent Meta’s operating systems, administrators, or other parts of its infrastructure from inspecting the underlying data.
The concept is similar to the privacy protections Meta has already introduced for private AI conversations. In May, the company launched Incognito Chat for Meta AI on WhatsApp and the Meta AI app, saying conversations were processed in a secure environment that Meta could not see.
For AI glasses, the same approach could eventually support private transcription, contextual conversations, and other personalised AI experiences. Meta has not yet specified exactly which features will use Private Processing when the technology reaches its glasses.
The company also says it is working to prevent requests from being easily connected to individual users. The system can use anonymous credentials and third-party relay services. At the same time, servers can be checked against publicly recorded information to verify that approved software is running before personal information is transferred.
Meta wants researchers to examine the system before launch
Despite the technical safeguards, Private Processing is not yet available on Meta’s AI glasses. Pritam Shah, Meta’s director of engineering, told WIRED that the underlying platform is already working. Still, the company is introducing details of the system before deploying more experiences so that security and privacy experts can examine it.
“What we are doing right now is sharing our excitement about this enhanced privacy platform,” Shah said. “We have the platform ready. We have it working. It’s not just a white paper and a blog. The reason why we are announcing it right now before putting more experiences on this platform is so that we can start working with the security community and the privacy community and a third-party auditor.”
Meta has not provided a firm release date for Private Processing on its glasses, nor has it confirmed which existing or future features will depend on the technology. Shah said the examples being discussed should not be interpreted as confirmed features.
The company plans to involve outside researchers in testing the system and expand its bug bounty programme to cover the technology. It also intends to give researchers access to relevant binaries and documentation so they can independently examine the privacy and security claims.
That approach comes as Meta continues to face questions about privacy and its AI glasses. The company has separately published information about safeguards for people around the wearer, including a capture LED that indicates when the camera is recording. Meta says covering or disabling the LED automatically disables the camera, and newer safeguards are designed to detect attempts to tamper with the light.
Privacy concerns extend beyond the person wearing the glasses
Private Processing could address one important part of the privacy debate: what happens to information belonging to the person using the glasses. It does not, however, answer every question surrounding wearable cameras and AI assistants.
AI glasses can see and hear aspects of the environment around the wearer, which means people nearby can potentially become part of an interaction without directly choosing to use the technology. Meta has acknowledged the need to consider both the wearer and other people when developing its AI glasses. She says its capture indicator is one of several measures designed to make recording more visible.
The issue has become more significant as Meta expands the capabilities of its glasses. The company’s latest AI systems are designed to understand visual information and provide more contextual assistance, while its broader glasses strategy includes bringing its Muse AI agent to the platform. Meta has also introduced new camera-free Ray-Ban Meta Audio glasses, alongside its next-generation camera-equipped products.
Meta’s privacy plans also arrive amid continuing scrutiny of its use of facial-recognition technology. A proposed class-action lawsuit filed in the US in September alleges that Meta used Facebook and Instagram photographs to develop its unreleased NameTag facial-recognition system and to train AI models. Meta has disputed the allegations and said it has not created a universal facial database or released NameTag to consumers.
The planned Private Processing system therefore represents a technical attempt to address concerns about how personal information is handled when AI glasses need cloud computing. Its effectiveness will ultimately depend not only on Meta’s architecture but also on independent testing, implementation and the specific features that eventually use it.
Meta’s decision to expose the technology to researchers before deploying it widely could allow outside experts to examine whether the promised separation between company and user data works as intended. Until the system is available and independently assessed, however, the privacy protections remain a combination of technical design and Meta’s stated plans rather than a feature that users can currently activate.
