Technology media The Information recently reported on new developments regarding artificial intelligence (AI) in the upcoming OS 27. Apple is actively exploring local edge AI solutions while ensuring user privacy, using Google's Gemini model to train a lighter AI model. This "distillation" technique allows the smaller model to achieve near-performance of the larger model, significantly reducing processing costs.

According to reports, Apple's AI team is working to use Google's Gemini model to "teach" a lightweight version suitable for devices. This design will ensure faster response times when users interact with Siri, while protecting user data privacy, in line with Apple's long-standing emphasis on prioritizing privacy.

In terms of cloud processing, since the full Gemini model has tens of thousands of parameters, the computing power required is extremely high, and Apple's Private Compute infrastructure cannot fully support it. Therefore, Apple has decided to transfer some of Siri's requests in iOS 27 to Google Cloud and use the licensed Gemini model for processing. Although the processing speed may decrease in the cloud, this will greatly enhance the security of sensitive information.

To further protect privacy, in recent weeks, Apple has approved the adoption of NVIDIA's "confidential computing" technology, which can encrypt data during processing. This means that even when processed in the cloud, user information can be kept more secure.

In pursuit of a more powerful AI, Apple is also actively seeking startups to help reduce the size of AI models. It once considered acquiring Liquid AI, a company focused on running AI models locally. Through these efforts, Apple hopes to find a balance between user experience and security.