Siri AI is powered by Gemini models, but is not Gemini – what does that mean?
Apple says Siri and Apple Intelligence use Google Gemini-derived models, but the assistant is still Apple's own system with different data and privacy controls.
Intelligence analysis by GPT-5.4 Mini
The piece argues Apple has built Siri AI from Gemini foundations without turning Siri into Gemini Assistant. Four models run on Apple Silicon or PCC, while the biggest model runs on Google servers under Apple’s privacy architecture.
Apple seems to have built Siri like a house using some bricks from Google, then changed the rooms, locks, and furniture to make it its own. Most of the work stays on the iPhone or Apple servers, but the biggest brain still has one foot in Google’s world.
Analysis
What Apple is saying
Apple has been careful to distinguish Siri AI from Google’s Gemini Assistant. The article says both systems may use Gemini models, but they are separate assistants with different product identities and different surrounding infrastructure.
How the models appear to work
The article points to Apple’s third-generation Apple Foundation Models, described as a family of five foundation models “custom-built in collaboration with Google.” Based on Apple’s wording and Jason Snell’s analysis, four of those models likely run on Apple Silicon and are customized for Apple’s needs. The fifth, and most capable, model appears to run on Google servers, though still as part of Apple’s own system.
The key point is that Apple does not seem to be using Google Search or Google’s knowledge graph directly inside Siri AI. Instead, the system uses Apple’s own data, weights, guardrails, and training pipeline, while borrowing Gemini frontier models as a foundation and refinement tool.
Privacy and trust
The privacy story is split across the model sizes. Two models run fully on-device, so the article says no data leaves the phone. Two more run on Apple’s Private Cloud Compute servers, where Apple says data is not retained or exposed, and that researchers can verify the system independently.
The open question is the largest model running on Google servers. Apple says PCC protections still apply there, and its security blog says the same safeguards remain in place. The article’s author treats that as the least proven part of the setup: not a claim of bad faith, but an uncharted deployment that may still hide unknown weaknesses.
Bottom line
The story’s main takeaway is that Siri AI is not a rebranded Gemini product. It is Apple’s own assistant architecture, apparently built from Gemini-derived foundations and wrapped in Apple’s privacy and device-first design.
Key points
- Apple says Siri AI is not simply Gemini Assistant in disguise.
- Four of the five Apple Foundation Models appear to be custom versions for Apple Silicon.
- Apple says two models run on-device and two more run on Private Cloud Compute.
- The largest model appears to run on Google servers but under Apple’s PCC rules.
- The article sees the Google-hosted model as the main trust and security question.
If Apple’s setup works as described, Siri could gain stronger AI abilities without giving up Apple’s privacy-first approach. The mix of on-device processing, Private Cloud Compute, and independent verification could make the system more trusted than a typical cloud-only assistant.
The riskiest part is the largest model running on Google servers, which the article says is still relatively uncharted territory. If Apple’s PCC protections do not hold up there as well as they do on Apple’s own servers, the privacy story could weaken.
