It was 2 a.m. when I opened a demo of the new Siri and felt a small, uncomfortable friction—responses that needed a cloud to finish the thought. You can hear the promise in Apple’s presentation, but you can also hear the quieter hum of someone else’s servers in the background. I want to tell you where that hum might be coming from, and why it matters to anyone who treats privacy as more than a marketing line.
I’m going to be blunt: Apple has been oddly quiet about where your Siri chatbot data will live and be processed. You and I both know the company built a reputation on doing sensitive work on-device. Now, anonymous reporting and a Bloomberg piece by Mark Gurman suggest Apple may be leaning on Google’s cloud to run parts of the new Siri. That’s a change with real implications for trust.
A software engineer on a beta call noticed Siri routing requests off-device.
On a recent call I sat in, an engineer paused and said, “Did that just leave the iPhone?” It’s a moment that captures the core tension: Apple’s marketing talks about Private Cloud Compute and Apple Intelligence, but the plumbing may include Google Cloud.
Gurman reports that Apple struck a deal to access Google Foundation Models, and the company will rely on Google’s cloud infrastructure for parts of the assistant. Apple’s public messaging so far stresses privacy protections and on-device processing, yet the company hasn’t updated its Foundation Models documentation to reflect the Google link. That mismatch matters because infrastructure shapes threat models more than slogans do.
Where will Siri store my data?
Apple says it will keep privacy central and use its Private Cloud Compute, but Gurman’s reporting suggests a hybrid: some processing on your device, and some routed through Google’s systems. Imagine your data moving between locked rooms in a house and a neighboring house where the locks are not exactly the same—your control changes with the address.
A roommate on a design team complained about limits placed on conversational memory.
One product designer told me the team is testing options that let users choose how long Siri retains conversation history—30 days, a year, or forever. That sounds like control, but control only matters if you can verify it. Apple plans to argue its memory model will be narrower than rivals’, but the company hasn’t published the cryptographic receipts yet.
Will Apple send Siri data to Google?
Reportedly, yes—at least for certain workloads. Gurman notes Apple is not saying it will adopt Google’s exact chips or security, but it may let Google handle part of the security stack. That’s a subtle but meaningful drift: sovereignty over your signals is weaker when they pass through someone else’s data centers.
A privacy researcher opened a dataset and found metadata pointing to third-party services.
That discovery mirrors a larger pattern: machine learning needs scale, and scale often rides on large cloud providers. Apple’s AI chief departure and the pace of development probably pushed the company toward a pragmatic choice—renting capacity instead of building it all. This is like borrowing a charging cable from a stranger when your battery is dying; it works, but you’d prefer to keep your own power.
How long will Siri keep conversations?
According to the reporting, users will be offered time-based options for retention. Apple plans to limit memories and impose restrictions on what persists. Still, many will not be satisfied until Apple publishes technical guarantees that show where and how those memories are stored, encrypted, and deleted.
Here are the things I’d watch for at WWDC (June 8): a detailed architecture diagram showing which services stay on-device, the exact contract terms with Google Cloud, and a clear audit trail for how memories are retained or pruned. You should ask for proof, not promises. When a company moves sensitive work to third-party infrastructure, you’ve traded a layer of convenience for a change in the risk profile.
Apple’s privacy reputation isn’t a logo—it’s a promise baked into design choices and cryptography. If those choices start routing trusted work through other vendors, the promise changes. I’m not saying Apple will betray users, but I am saying you have a right to know whether your assistant is storing secrets on a device Apple controls or on servers one of Apple’s biggest rivals runs.
Mark Gurman, Apple Intelligence, Google Foundation Models, Private Cloud Compute, John Giannandrea, OpenAI, and ChatGPT are names you’ll see across the conversation. I’ll be watching how Apple frames this at WWDC and what the documentation actually shows about data flows, encryption, and deletion policies—because what’s left unsaid today becomes the default tomorrow.
Trust is a fragile thing: hand it to the company that built a reputation around privacy, and you expect a locked vault; hand it to a vendor you don’t control, and the lock looks different. Are you ready to accept that different lock, or do you want evidence before you decide?