Apple is now displaying permission prompts when users invoke AI features that route data through Google’s servers, a move that treats Google’s infrastructure similarly to how the company handles third-party AI services like ChatGPT. Shape generation in iWork on iOS 26 and comparable AI features in Freeform on iOS 27 trigger these warnings, making the data flow explicit rather than silent.
The warnings arrive as part of a broader shift in Apple’s AI strategy. The company announced in June 2026 that its next generation of Apple Intelligence would be powered partly by custom foundation models trained in collaboration with Google, using Google’s cloud infrastructure and Gemini architecture as sources for the collaboration.
Apple has structured the Google collaboration to maintain clear boundaries. The company uses none of the Gemini models deployed to Google’s customers, none of Google’s client-side code, and no Google Search infrastructure as the knowledge backbone. Instead, Apple built custom foundation models informed by Gemini’s architecture, then routes queries through an orchestrator that directs each request to the appropriate model based on complexity and personal context required.
This federated approach balances two extremes, processing everything on-device (which limits capability) and offloading everything to a third party (which risks privacy). Requests that require personal context or are too complex for on-device handling go to Apple’s servers running Private Cloud Compute, and only when those servers cannot handle the task does data route to Google’s infrastructure.
Apple is relying on Nvidia’s confidential compute feature to protect data as it moves through Google’s systems. This Nvidia technology encrypts both data and AI models during processing, preventing even Google’s infrastructure teams from accessing the raw information. The encryption carries a performance cost, queries process slightly slower, but Apple views this as an acceptable tradeoff for maintaining privacy guarantees at Google’s scale.
The permission prompts signal that explicit disclosures will satisfy privacy-conscious users better than hidden infrastructure decisions. Rather than quietly routing requests and hoping users never notice, Apple is making the data flow visible. This differs from how the company handled Private Cloud Compute initially, which functioned without user-facing warnings because data never left Apple’s ecosystem.
Whether users accept this transparency or view it as a concession to Google’s scale remains an open question. Apple’s privacy reputation has been built partly on keeping data off third-party clouds entirely. Prompts make that claim harder to sustain, even if encryption protects the underlying data.
