Apple’s artificial intelligence servers will use ‘confidential computing’ techniques to process user data while maintaining privacy

At WWDC, Apple will unveil its AI strategy as part of iOS 18 and its other operating system announcements. As previously reported by Bloomberg and others, the company plans to exploit a combination of on-device and server processing.

However, processing of user data on Apple’s server infrastructures raises privacy concerns, especially when Apple has been touting on-device approach for so many years. Today, The Information explains Apple has what it thinks is a solution to provide rich AI processing abilities in its cloud while maintaining strict privacy standards.

The Information report describes that Apple intends to employ confidential computing techniques, which will enable “black box processing”.

Typically, cloud services encrypt data only on disk, when it is being stored. But the data has to be decrypted into memory in order to be processed or transformed on the server.

What The Information claims is Apple has found a way to process user data in such a way that it remains private throughout. It says Apple has upscaled its Secure Enclave designs to enable such a programming model. Bloomberg previously mentioned the relationship to the Secure Enclave with the Apple Chips in Data Centers (ACDC) project.

The Information says there is still potential weaknesses if hackers assumed physical access to the Apple server hardware. But overall, the approach is far more secure than anything Apple’s rivals are doing in the AI space. For instance, the system is so secure that Apple should be able to tell law enforcement that it does not have access to the information, and won’t be able to provide any user data in the case of subpoena or government inquiries.

Interestingly, it seems work on this confidential computing initiative predates the AI boom, with The Information saying that work inside Apple has been ongoing for at least three years. While it has short-term uses for enabling some AI functionality in iOS 18, Apple also has longer-term plans for the technology.

For instance, the report says in the future Apple could create lightweight wearable devices that don’t require powerful chips, as they could offload their processing to Apple’s backend.

However, the exact details of how this will work are still murky. The Information said it remains unclear how Apple will be able to preserve the security model, when a singular chip in a data center is (naturally) running requests from many users simultaneously.

There is also little visibility into how much Apple Silicon cloud computing scale Apple has at its disposal right now. As such, it’s possible the initial launch will be limited. Previous reports have indicated that Apple only just started putting M2 Ultra chips in its cloud, with plans to move to M4 deployments by the end of the year.

You can expect to hear more about all of this in a matter of weeks. Apple’s WWDC conference kicks off on June 10, in which Apple will official unveil its AI strategy for the first time.