Apple has integrated Google’s Gemini models into Siri AI, but the two systems remain separate. While both leverage Gemini’s foundation, Apple customized the models for its devices and privacy needs. Four of the five models run on-device, ensuring data never leaves your iPhone. The remaining model operates on Google servers via Private Cloud Compute, but Apple claims identical security protocols apply. This hybrid approach balances innovation with user privacy concerns.

The partnership stems from Apple’s third-gen Apple Foundation Models, four of which are modified Gemini versions optimized for Apple Silicon. These models use proprietary data and reinforcement learning, refining outputs from Gemini’s frontier models. The fifth, most advanced model runs on Google’s dedicated servers with NVidia GPUs, though Apple insists user privacy is preserved through its PCC architecture. Security researchers can independently verify that no data is retained or shared with Google, even in this cloud-based setup. The key distinction is that Siri’s processing—whether on-device or in the cloud—remains entirely under Apple’s control.

This integration matters because it shows Apple’s willingness to collaborate with Google while maintaining strict privacy boundaries. Unlike Gemini Assistant on Android, which relies on Google’s ecosystem, Siri’s use of Gemini models is tailored to Apple’s hardware and security frameworks. The most powerful model’s reliance on Google servers introduces uncertainty, as PCC implementation here is uncharted territory. While Apple asserts it meets all PCC requirements—stateless computation, no privileged access—the lack of public audits for Google’s servers raises questions. For now, users must trust that Apple’s safeguards hold, even when processing occurs outside its infrastructure. The move also highlights how tech giants navigate partnerships without compromising core principles, at least in theory.