Five new Apple Foundation Models and the privacy promise

Apple's new AI systems are private and secure without a drop of Google on your device

WWDC 2026 may have been a different format, but we still got a ton of new features and design changes on top of the revamped AI. It's a busy week at AppleInsider, so I don't have long to write here.

First, I wanted to take a bit of a victory lap. As I wrote yesterday, Apple's upgraded Apple Foundation Models that run on your devices and in Private Cloud Compute don't have a drop of Gemini in them.

There seems to be some confusion about what I'm saying here, even though I'm using plain English, but here's what I shared to elaborate my point in the AppleInsider forums.

I believe the whole point I'm making, and people's concerns are, whether or not interacting with Apple Foundation Models has any connection to Google, its models, its code, or its search engine.

I'll say it again: it does not.

However, that isn't me saying that Google wasn't involved in the development and training process in these new LLMs. In fact, version 1 of AFM was partially built using Tensor. I'd sound like a crazy person if I were suggesting that Google technology isn't in the stack, when it clearly is stated that it is.

However, the presence of Google's underlying server stacks, Nvidia's chips, and Apple's own technologies doesn't mean any one of those entities is getting your data or using it for AI training. That's what people are asking, and that's what the article describes when I say there isn't a drop of Gemini in AFM.

These are Apple models through and through. They're improved because of Apple's access to Google's core LLM specifically segmented for training Apple models in Apple hardware.

Their end result of the partnership is what we see – these five powerful models. But just as TSMC owning the fabrication process for Apple Silicon doesn't make them TSMC chips, Google having the more capable model to beef up Apple's offerings doesn't make it a Google model. The structure, behavior, function, guardrails, and feature set were all designed and implemented by Apple. Google's models reinforced and improved Apple's, and will continue to do so! This partnership isn't over!!

BUT, people can rest assured that the models installed on their devices aren't Google's in any part beyond the technological foundations. Google doesn't have any active role in how AFM functions nor how data is handled.

So yes, I'll say it again, there isn't a drop of Google in your AFM. They are wholly an Apple technology at the implementation stage.

I understand my phrasing could cause some confusion if your interpretation of Apple's partnership is that Apple's new models are basically retrained versions of Google's. But that's not what is happening here. You don't have to believe me either – Apple's executives explained, in detail, exactly how AFM is not using Gemini or Google in any part of the shipping models.

Google is present and an active partner with Apple. Its core LLM will absolutely influence how Apple's models will work! I'm not suggesting otherwise. The partnership enables Google's more capable models to increase the capability of Apple's.

That's a good thing!

The whole point I want to convey, and what I've been repeating for months, is a simple fact. Your data is yours when you interact with Apple products. Apple Foundation Models that run on device and in Private Cloud Compute servers are Apple's through and through.

There isn't a little Google gremlin living in your iPhone or in the cloud sifting through data. Nothing you do is going to Google, Nvidia, or even Apple in any way beyond how you're using the AI. Your data is encrypted, minimized, and tossed.

When you search world knowledge with Siri, it isn't performing a Google Search anymore. It's parsing its own database. When you generate an image, it's doing so with an Apple-made image generation tool. Not Google's.

Yes, Google plays a role in helping Apple improve its models, and I'm glad it is. Google is an industry leader in this space with an actual business model that will ensure it'll stick around long after the AI grift bubble pops. Google has a lot of ethical issues; its AI stole much of its knowledge, and the new Search is trying to destroy the open web. That's also true.

But Apple partnered with a leader in the space that it already had a relationship with. The end result is a series of models and continued improvements that will make using our Apple products better while maintaining privacy and security standards.

I'm quite happy with that outcome.

Edit: John Gruber over at Daring Fireball shared a rather succinct way of framing this that I'm happy with sharing here. The confusion here is how Apple and Google refer to Gemini as both the model and the agent. Once is consumer facing one is not. Apple got Google's LLM to help train AFM, but the end result is not a clone, inclusion, or derivative of Google's consumer-facing Gemini chatbot. Maybe that helps.