Very Apple-ish approach to AI catch up: wrap an external tool in a privacy architecture, embed into the OS and productize the orchestration layer.
It will be interesting to see if the Private Cloud Compute + on-device routing can make third-party model capabilities feel like a first-party system without leaking user context to the model provider.
If Apple handles the Google-Apple boundary right, this will be an elegant move on their part, otherwise it will feel like Apple Intelligence with a just a privacy-polished frontend for Gemini.
As someone who doesn't use Android, they showed a lot of integration into the apps, which I think is where the real magic happens, and it's not something I can do with any 3rd party chatbots today (that I'm aware of). I also don't know that I would trust the other 3rd parties with the access required to pull it off.
I'm interested in how it feels to use: whether there is any context leaking, as you mentioned, if it introduces latency, and whether there are any pricing implications? I know they weighed a variety of factors, including the smaller models, but cost had to be a big concern, too... I feel like Google is the only provider giving away so much AI inference for free.
This will further blur the picture about when and how consumers / employees are supposed to pay for AI services. For example, they showed consumer rather than coding tasks, but could you select five files and ask Siri to write a Python script or a small app? Will enterprises just disable Siri AI functionality, or will they be able to route it through their own AI auditing and providers?
Before you start driving, at stop lights, while waiting in lines, etc.
I don't know how it works on CarPlay but when I turn my car on I have a bunch of suggested addresses (home, work, parents, recent Maps searches, etc) that I just touch-to-go. Having to use voice every time you want to navigate not only sounds unnecessary, but cumbersome.
You don’t need to enable Gemini or voice assistant on android to use android auto. Some functionality is lost, of course, but you can still navigate and play music.
It’s the DMA regulation that forces Apple to give the same access as they have to other AI chat apps.
Once it leaves the device Apple does not know what those other ai chat apps will do with the gathered data.
> Siri AI is private by design and deeply integrated across Apple’s platforms using on-device processing and Private Cloud Compute, which extends the privacy and security of iPhone into the cloud. However, under EU regulators’ extreme interpretation of the DMA, Apple would have to give any virtual assistant direct access to users’ private data — and the ability to directly control other installed applications — as soon as Siri AI is made available in the EU, without the essential protections necessary to keep users and their data safe.
As a screen reader user, Apple does have a first party screen reader, VoiceOver, and does indeed not let you run a third party one. In fact, it does not work well even on the more open MacOS. So essentially it's VoiceOver or nothing. Luckily, especially on iOS, VoiceOver mostly works well.
I'm glad it works decently on iOS, at least. my mom has little central vision, and she struggles on iOS just using high contrast plus scaling plus magnifier. I think she has just enough vision to not absolutely need VoiceOver but it still makes using her phone a frustrating and tiring experience.
I don't think they're "trusting Google" with anything. It's a Google model run by Apple, just like you download a model from Hugging Face to run when you want to.
I don't want my apps that have AI implemented to be able to read my messages because Europe mandates feature parity. And that's just the tip of the iceberg. For Apple it means building all the APIs that probably already exist but this time to be requested by apps, which would be a huge attack surface, even Apple's own apps suffers from security breaches (like Message before the switch to closed container execution). AI breaks the separation of concerns, which can lead to disastrous consequences.
EU has great intentions, and of course, feature parity should be offered so that competition can exist, but I don't find it crazy that it is more complicated on a product like that. As tech people things are very obvious to us but we need to remember that we are talking about a product used by everyone.
It’s not clear how it is significantly different from allowing apps access to your contacts, calendar, photos, and so on. And Apple doesn’t say that they merely need more time to properly implement it, the claim that they are unable to implement it without compromising privacy and security. And the latter I don’t really see, with the proper set of permissions presented in the way users are already used to.
As an Apple user I feel more patronized than empowered here.
> It’s the DMA regulation that forces Apple to give the same access as they have to other AI chat apps.
But why can Tesla ship Grok to their cars in the EU without any problems? Why aren't they required to let me choose between Grok, OpenAI etc or even a custom endpoint?
> But why can Tesla ship Grok to their cars in the EU without any problems?
Simply because they are too small in user count. EU DMA, DSA etc. only apply at certain thresholds. Twitter for example falls under the scope, but Tesla is a distinct entity from Twitter and even if they were merged together, they would still be distinct services in the eye of the law.
I think there’s a case that Apple’s commitment to privacy here will increase participation by 3rd party developers.
For example, if I’m maintaining a secure chat app, I think I’d be more likely to adopt the APIs to share the chat messages with the system AI due to Apple’s promises that the data will either be processed On Device, or in their Private Compute Cloud.
If I instead believe that sharing the chat messages with the system AI would cause those messages to be sent to unknown-to-me other entities, I think I’d be less likely to participate in the new API.
This user might be okay with their data going to this other provider, but what about the people they’re messaging? I have a responsibility and a commitment to _all_ of my users to protect their data.
I might not be able to control what any specific user does with the data, but proactively writing the code that sends the chat messages to this other system is something that I have control over.
And that's exactly how it works for apps you download from the App Store. Apple even makes app publishers declare data collection and privacy practices on the App Store before you install apps.
It's clearly just Apple not wanting to further open up their platform to competition.
Since it's the user's device, not Apple's, EU correctly "interprets" this as the user has the right to do whatever they please, including installing third-party chat apps.
Apple are just bulshitters when it comes to actual users, and not their corporate definition of a user.
This is such a shallow take. There are obvious privacy and security tradeoffs here. The EU competition framework is good in many ways, but this is actually something I don’t think we have the regulatory frameworks in place to handle yet , or social norms and understanding about why giving any Tom dick and Harry root on all your data is a bad idea.
It’s paternalistic, but I agree with Apple that free for all access to this kind of data is not a great idea. Ironically, before this could work we’d actually need much more EU style data regulation, and more consistently enforced.
Apple uses "privacy and security" as a cudgel to prevent anyone from breaking into the vendor lock in. To the point that EU actually had to explicitly tell Apple what to do [1], as Apple delayed features, made them extremely hard or convoluted for third-parties to use, and pulled every trick out of the malicious compliance manual.
This whole virtual assistants thing will drag on for another several years.
Unless Apple proves otherwise I'm more inclined to believe they're either 1. Using this to try and shape the DMA in their own interest (definitely not their users' interest) or 2. Doing something with the data that would not be allowed in the EU (also not in their users' interest at all) or both.
From the link "Nvidia has its own “confidential computing” feature that encrypts data as it’s being processed, which will be used with other privacy and security measures to protect user data"
They say it's because the EU's DMA would require them to open up device data to third-party assistants, and they'd no longer be able to guarantee user privacy.
I don't see what the issue is. The user could then select Apple (or Mistral) for strong privacy or another provider for customers that don't care.
I primarily want Apple to provide extension points so that I can select my own provider, just as I can choose where to host my mail or install another app as my instant messenger.
Sure, I could install another provider's app, but it wouldn't have the same integrations, similar to how an instant messaging app would be less useful if notifications were limited to iMessage.
Nothing in the law requires the pop up. It definitely doesn’t require the obnoxious bullshit that most companies put up (aka the dark pattern to get you to agree to every unreasonable part of their terms just to read the page).
The alternative would be to just stop invasive tracking and add the cookie when it’s actually needed.
The smell is that Apple doesn't want to give the same level of access to third-party AI assistants that Siri will get.
For what it's worth, Apple claimed they proposed an "equivalent access" framework with some kind of "trusted agent framework" approach, but that it was shot down by the EU. I suspect it was way more inconvenient for third-party developers than Apple lets on.
This has been the case for quite a while. Like Reminders - You can't replace the phrase "Siri, remind me to ___" with a third party app. I'm surprised the EU lets them ship Reminders there.
> The company reiterated that Apple Intelligence relies on on-device processing and Private Cloud Compute, with a promise that user data is only used to execute the immediate request and is not accessible to Apple or third parties. Apple added that outside experts can verify those privacy guarantees "at any time."
Yes, the "Apple needs to look at your data to do this, but we don't have any way to look at the data if we wanted to". That's impossible, unless they open souce iOS and let people take control over their devices, and let people self host inference, so people can check that there is no network traffic. If it is as they say, they could let people host it without any downsides.
It's strange to me that Apple would choose to disadvantage themselves by selecting Google as their provider as opposed to, say, Anthropic or even OpenAI. Doesn't this mean they'll struggle more to differentiate themselves from the assistant on Android phones? Thinking more cynically, couldn't Google, if they wanted, feed Apple an inferior version of Gemini, ensuring they stay ahead?
As the consumer, this just sucks because it means no matter which phone platform you choose, you're getting the same thing underneath, and there's no way to avoid it (besides not using an assistant entirely, which I recognize a lot of people do, myself included).
I'd wager that for 99.9% of "Apple Intelligence" tasks, Google's models perform just as well as other frontier labs. Google also has done more work on getting LLMs running on edge devices compared to anthropic and openAI.
The source also says
> The new architecture centers on Apple Foundation Models co-developed with Google, which Apple says are adapted to run both on-device and on servers through its existing Private Cloud Compute infrastructure
Which could mean Google and Apple have trained some custom models, probably the on-device ones, specifically tailored towards Apple's hardware.
That's not the point of OP. The point of OP is that Google is a direct competitor of Apple for phones / OS, so them giving the "key of the house" to Google is risky
What indicates they’ve given google anything other than a truckload of cash? There’s zero data sharing in the arrangement and the stuff is running on apple devices and in Apples Private Compute.
Yes, but you get less risk in other areas. Google is a long standing publicly traded company. Apple knows that they know their stuff, and that they’re gonna stick around. Anthropic and openAI are new kids on the block when it comes to software as a whole, and that’s a risk.
Apple and Google have been dealing with each other for quite a long time. My guess is that they want to replicate the relationship they have with Safari, where Apple provides the users and Google provides the search engine (and money).
Last year the announced they were working with OpenAI. It looks like this went nowhere, so it's not really surprising to see them try someone else.
They already have a very codependent relationship because of their revenue share over putting Google search up front in iPhones so I doubt either party would put that at risk.
I see it differently… Apple has chosen to treat the model as a commodity. By making Gemini an implementation detail, they leave the door open to swap it out for Anthropic or OpenAI without end-users even knowing or caring. So I think they're creating leverage in any future negotiation.
And at the same time they claim it can't be rolled out in Europe because the DMA would force them to allow selection of other AI providers? Which would not even be true if this was an Apple product built on the models of Google, just like the DMA does not force them to pick a different datacenter of office cleaning provider.
Take it with a grain of salt but I don't think it's other AI providers that Apple is upset about. The DMA would require users to be able install any openclaw like thing onto their device with access to everything that Siri can access today. There are all sorts of arguments to be made here but I can understand why Apple feels this way and wants to offer a good experience here.
Looks like the problem with DMA has more to do with giving full access to other providers outside of the on-device + private cloud compute architecture. The way I interpret the newsroom post [1], Apple doesn't want to give third-party providers full access to user data when the third-party providers cannot run on private cloud compute for privacy reasons, but the EU wants them to offer the choice anyway.
> they claim it can't be rolled out in Europe because the DMA would force them to allow selection of other AI providers
They don't claim that. All they said is "later in the EU as we look into privacy and security" after spending two hours saying how private and secure everything is.
DMA would force them to allow usage from other apps than their own and other assitants than Siri, especially for on-device models.
> I see it differently… Apple has chosen to treat the model as a commodity
It didn't work out well for Yahoo.com. It turned out that Google Search was the value and yahoo.com just skin around it. It might be the same for Apple. Gemini is the valuable part, what particular device you use it on matters less.
Google search was leaps and bounds better than any other search engine when it came along and dominated. Yahoo couldn’t build their own, and nobody else they could buy from compared.
As i understand it, no LLM is miles ahead of the others right now, especially when it comes to simple agentic stuff. Hell, Qwen3.6-35B-A3 quantized to 3bits running on an 8 year old consumer GPU handles most agentic stuff fine, if a bit slow.
Differences in LLMs boil down to mostly the harness and the compute to run the models. Even for high complexity tasks like coding, the differences between openai, anthropic, google, and the bigger qwen models aren’t that dramatic.
Bingo. They see this as the future commodity it will be. Customers will choose AI providers much way they choose a car: taste, price, a few other factors.
And to your point, Google has a massive balance sheet, produces their own AI chips, and is not going anywhere anytime soon.
> It's strange to me that Apple would choose to disadvantage themselves by selecting Google as their provider as opposed to, say, Anthropic or even OpenAI. Doesn't this mean they'll struggle more to differentiate themselves from the assistant on Android phones? Thinking more cynically, couldn't Google, if they wanted, feed Apple an inferior version of Gemini, ensuring they stay ahead?
Is it really all that difference from Apple defaulting to Google's search engine?
Depends on which way the money is flowing. Google pays Apple for default search engine. Is Google paying Apple for using Gemini? That feels like a much heavier investment if they are
> It's strange to me that Apple would choose to disadvantage themselves
How exactly are they disadvantaging themselves? Perhaps expand on that opinion a bit, and the data/assumptions you're making in forming it.
My naive assumption is that they're going to do what everyone is doing: make tooling that lets you swap in any model.
I don't think it's fair to assume incompetence, on their part. I think it's much safer to assume they're doing what's best for them, and it's very clear to them what's best (money, support, etc).
At the time Apple made this decision there wasn’t as strong of a difference in model quality between Google, anthropic, and OpenAI as there is now. Also Apple definitely burnt some bridges with OpenAI on the agreement they made together a year earlier.
Especially if you are not asking siri to code stuff. If your use case is still "personal context" with "conversation", even if there is a 9/10 difference, the users may see their abilities going from 3/10 presently to 6/10 or 7/10, and still a massive upgrade for most of them. I think therapy and scientific research may remain in the domain of frontier chat interfaces for another year.
OpenAI and Anthropic don't make small models. Google happens to already have a billion devices that would benefit from small models, so they made one.
Google basically gets 1 billion per year for free*.
Gemini is better than either at multi modal, google also has their tensor processor stuff with ridiculously high T/s output they need for acceptable UX
Yet? Maybe they are but embedding adds or paid signal amplification/probability tweaking has already been floated on the market and may already be a product.
It doesn't matter if it's classic search or LLM. They can monetize tracking information as easily as they can sell ads. They'll have fast cheap custom-built assistant models that run on device by default, keeping things profitable. In time they'll likely double-dip again by injecting product placement in results.
Search has ads but Siri doesn't. And when Apple puts ads in Siri they won't be Google ads. I don't think Google benefits from this deal enough to be worth paying.
Maybe. Search ads likely make Google more money than they pay apple. For AI, Google currently loses money. If they eventually make money via ads, then sure. Else, apple will have to pay them
This is the most likely explanation. Apple manufacturers some of the best inference silicon on earth. Apache licensed models are already 1000x smarter than siri and strongly outperform anthropic, openai and google in the 8-128GiB of RAM range. The article says Apple can run this stuff on customers’ hardware, so that’s the range of model sizes that actually matter.
> Apache licensed models are already 1000x smarter than siri and strongly outperform anthropic, openai and google in the 8-128GiB of RAM range
smart is a weird term, gemma4 is an amazing omni model better than qwen3.6 for non coding tasks (as for all Gemini models). For Apple Intelligence gemma4 makes a lot more sense.
A functioning FTC would not allow this. Insane that there will be no competition for integrated smartphone AI because the existing duopoly at the OS level has agreed to team up.
It may be a clever move. By using the same models as android (contractually?), they can compete on the user experience which they typically handle better than android phone providers.
Anthropic or OpenAI have no foothold into the mobile market. Google has integrated a shitton of AI functionality into their latest Pixel phones. That’s what would scare me if I was Apple and worried that if AI prevails this could steal some market share from me. The other two are irrelevant in this context.
Didn't they famously have a search deal with Google (that they were also ultimately fined for - in the EU at least)? So there's definitely precedent with Google as a "partner".
I'm speculating, but it's likely that Google is the only provider who is willing to adhere to the inference compute requirements that Apple sets out for their foundation models. They are, after all, the only provider that will let you host their FMs in your data center.
What part of using Gemini do you think is a disadvantage?
Also important to remember how immature OpenAI and Anthropic are as companies. It would be a huge technical, legal, and reputational risk to commit to using them.
> Apple Intelligence is designed to protect your privacy at every step. It’s integrated into the core of your iPhone, iPad, and Mac through on-device processing.
The agent harness matters just as much as the AI model. Using Hermes or OpenClaw feels like night vs day when using OpenAI’s apps even when using the same exact model.
You can even see difference in agent harnesses using the same model in the same company if you compare Gemini CLI with AntiGrav. They are different experiences.
I’m pretty sure Apple’s agent harness will be drastically different from Google’s even with the same model
The differentiation is the integration, not the model itself which is mostly fungible. And afaik Apple is running these models on their own compute, so I don't think google can pull a bait and switch.
I would love to learn more about what's actually powering Apple Intelligence now. Are they using flagship Gemini models behind their own prompts? Fine-tuning? Pre-training their own models based on Gemini?
Is there a meaningful distinction between the Gemini-powered models and Apple Foundation Models? Does that distinction vary for on-device vs hosted models? Are some models running on Apple's Private Cloud Compute and others running on Google iron?
Edit: they elaborated significantly in a "keynote tech-talk": [0]
According to Apple, there are five models:
On-Device
- AFM Core: Dense architecture; the standard next-gen on-device model
- AFM Core Advanced: Sparse architecture, natively multimodal; enables features like image understanding and expressive voices
Private Cloud Compute
- AFM Cloud: Workhorse server model optimized for latency and cost
- AFM Cloud Image: Image generation and editing
- AFM Cloud Pro: Most capable model, Gemini frontier-level quality, for complex reasoning and agentic tasks; runs on NVIDIA GPUs in Google's cloud under Apple's PCC privacy guarantees
Everything excluding Cloud Pro are custom models running on Apple Silicon, "refined" using Google Gemini. About Cloud Pro, they say "this is our most capable model with quality similar to Gemini frontier models." So I might read between the lines and say this is a wrapped Gemini.
Local is probably similar to Gemma e4b you can get right now on Google Edge Gallery (the ios and Android app). Guessing that the more powerful version that will only work on the 12gb ram devices will be something unreleased that is similar but a bit larger
Google also awhile back announced being able to run full Gemini by leasing / renting hardware in your own datacenters so companies can train or access data without needing to send things to their datacenters. Nvidia based. Guessing Private Compute might just be Apple leasing a ton of those?
The public version of Gemini is ridiculous. At least half their search "answers" are just wrong. If you then start a follow up chat the answers change but usually still half wrong.
Search would be better without the added AI hallucinations above it. If I want an AI answer I'll go and ask Claude, the quality difference is huge.
> The public version of Gemini is ridiculous. At least half their search "answers" are just wrong.
That's not Gemini, that's AI Mode (in Search), they're different products built by fairly different part of Google (actually one is built by Deepmind).
(I don't think it's much comparable to https://gemini.google.com/app at least in the past you'd get very different results)
It has to be really because think of how fast it has to come up with an answer (ie time for a regular google query) and the immense scale of billions of people querying it many times a day, all for free.
It is weird and disturbing that Apple has no native AI capability.
This is one of the most cash rich companies in the world and it has failed to have any position in the most critical technology development perhaps ever.
It's a clear signal that Apple became the most incredible operational/execution company under Tim Cook, but lost its innovation leadership.
I take it as a signal they don't see any of their value being provided by the models. They're strong point was never frontier technologies. It's always been the delivery of the technology.
Really I don’t think this is a strong take at all. If anything this has positioned them extremely, extremely well for when the bubble bursts and they can go with the winner to provide reasonable capabilities.
>The company reiterated that Apple Intelligence relies on on-device processing and Private Cloud Compute, with a promise that user data is only used to execute the immediate request and is not accessible to Apple or third parties. Apple added that outside experts can verify those privacy guarantees "at any time."
> Apple added that outside experts can verify those privacy guarantees "at any time."
To my understanding, these "outside experts" have to go through a vetting process by Apple first. There are no publicly available audits of the infrastructure planned right now.
This move kind of reminds me of the original iPhone with google maps. You're competing with google, but you're using their infrastructure. Why wouldn't they just go with another provider like OpenAI or Anthropic?
To be fair, with the original iPhone they weren't competing. Google did backend web stuff well, so Apple partnered with them and made nice local apps that were fed by Google's cloud data. The same was true for the YouTube app.
After Google saw the iPhone (before the public), they pivoted their vision for Android (it was originally on blackberry-style hardware), and that's when the "thermonuclear war" started. Kind of interesting Steve Jobs would have showed the iPhone off like that, when something very similar happened with Bill Gates, which prompted the development of Windows (according to Jobs).
19 years later, it's probably time to be pragmatic again. If Apple isn't able to deliver on some of these AI integrations into the OS, they risk losing users to Android. If they have to pay someone for a model, they might as well choose the one they think is going to be best for their users. This keeps existing iPhone users on iPhone, and may pull over some Android users looking for the same features, but with better privacy. That seems like a win for Apple. To pay OpenAI instead of Google would just be spite at this point. Maybe well deserved, but the leadership has all changed over in the last two decades, so they'd be fighting old wars. Though I think they should still think twice before showing Google anything that hasn't yet been released.
Given that they had originally selected OpenAI for Siri, and that deal fell through, I would guess something about their relationship with OpenAI fell through. Maybe OpenAI wouldn't let Apple run their model on Apple's servers.
Google is almost closer to a conglomerate than a coherent horizontally integrated corporation. The individual parts of Google are like Fortune500 companies themselves, and tend to act in their own interest.
OpenAI or Anthropic are not anywhere as well funded as Google. Apple already has everyone in their pocket via the ecosystem, they just have to not crap the bed. They value stability over the competitive component here.
Most Android devices aren't even Google products, and for much of Android's existence, none of them were. Android is a way to get ads in front of people... as is their partnerships with Apple.
Google apps are the most downloaded apps in the Apple App Store already. This reminds me of the original Apple Maps, which was just a front end for Google Maps.
Google sold Apple the ability to run certain Gemini models on Apple's data-center hardware, using Google's orchestration layer. Apple hooks into that not dissimilar from an API-provider, and then builds everything upstream.
Meaning the system prompt(s), harness, entry and exit points, and skills. So the product is still "Siri AI", because of all the stuff that takes it from a raw infrastructure concern upon up into a "product" is Apple's responsibility.
Google are "okay with that" because Apple pays them $1B a year, per press reports, to be.
Any business can do this now actually - you just need to lease/rent hardware through a Google partner and you can run an Nvidia based server in your own datacenter running (supposedly) the latest full Gemini models.
I don't think you conceptualize Google's game plan. all these companies care about is b2b contracts so they can inflate their balance sheets because when it's digital, it doesn't have to actually exist for it to "make money"
What a blunder, they resisted AI for like 2 years when it was all the buzz, and now when the bubble is about to bust and every user has AI fatigue they decide to finally dip into the fad?
Before it was as if avoiding AI was a conscious design decision, and if there was an AI crash, Apple would be the only survivor left. Now it feels like they weren't in on the meme out of incompetence and are now late to the party.
No one can know what Jobs' stance would be, but I like to think he would be anti-slop
what i think why they are too much relying on Google is coz they are way much towards making models open source and launching more much better models to public as if in future apple part way from google they might still have much better models to rely on and if we see the history google has been partnered with Apple since the launch of first Iphone
I honestly don’t understand how anyone can believe that Apple is limiting user options for privacy reasons, rather than trying to maintain an unfair advantage over other vendors.
I’m not saying people who hold this view are being dishonest at all. But sometimes, to me, brands like Harley-Davidson or Apple seem closer to a cult than to a typical corporation.
Because privacy is actually a big feature, many people are skeptical about AI and the big model providers and don't trust them. They're less skeptical about on-device AI and so Apple is pushing that and making privacy a core feature of their online offering as well.
I probably wouldn't use it without that. It's one thing sending my shitty code to be trained on, and another thing entirely to give these companies access to my personal life and information.
Sounds like Apple offered to implement that, but it didn’t fit the EU’s requirements.
I suspect if you paid apple enough money, and were willing to prove that your personal Private Cloud Compute did meet their requirements, it wouldn’t be impossible.
When you have to image a highly irrational reason to explain why groups of people do the things they do, there's a decent chance you just don't understand their perspective. They may be acting reasonably rationally from their own perspective. (As you said yourself: "I honestly don't understand...")
Apple could have done something like bedrock and used a SOTA model but instead they are fiddling with local models or whatever.
Also I have seen that Apple has some strange lust towards image generation as if that's what people really want. I have this slop image generation thing on my phone and it is useless.
Here's what I want: natural language interaction to achieve complex workflows in iPhone. Example: find the cheapest way to go from A to B and book it using the Deutsche Bahn Train app.
I don't disagree that local Gemini is better but if you've tried it in iPhone, it is slow, hallucinates and overheats the phone. For anything slightly non-trivial like the workflow example I gave, I think it will be close to useless.
Why is Apple providing people with a photorealistic deepfake generator so they can participate in dressing down women, digital blackface, and god knows what else? This is crossing a line, and simply saying "well other big tech companies crossed it first!" is not an excuse.
How so? Can you not create all sorts of unethical imagery with the photo editing apps on the App Store? And do they not make the production of unethical imagery much easier and accessible? Which part of my comment is ridiculous and absurd?
It will be interesting to see if the Private Cloud Compute + on-device routing can make third-party model capabilities feel like a first-party system without leaking user context to the model provider.
If Apple handles the Google-Apple boundary right, this will be an elegant move on their part, otherwise it will feel like Apple Intelligence with a just a privacy-polished frontend for Gemini.
I'd use this.
I'd rather have strong privacy guarantees, but this is still good.
This will further blur the picture about when and how consumers / employees are supposed to pay for AI services. For example, they showed consumer rather than coding tasks, but could you select five files and ask Siri to write a Python script or a small app? Will enterprises just disable Siri AI functionality, or will they be able to route it through their own AI auditing and providers?
I don't know how it works on CarPlay but when I turn my car on I have a bunch of suggested addresses (home, work, parents, recent Maps searches, etc) that I just touch-to-go. Having to use voice every time you want to navigate not only sounds unnecessary, but cumbersome.
CarPlay does not work at all if you have not enabled Siri. As in it won’t even connect.
Once it leaves the device Apple does not know what those other ai chat apps will do with the gathered data.
> Siri AI is private by design and deeply integrated across Apple’s platforms using on-device processing and Private Cloud Compute, which extends the privacy and security of iPhone into the cloud. However, under EU regulators’ extreme interpretation of the DMA, Apple would have to give any virtual assistant direct access to users’ private data — and the ability to directly control other installed applications — as soon as Siri AI is made available in the EU, without the essential protections necessary to keep users and their data safe.
https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-de...
- Apple has powerful capabilities in iOS to enable Siri AI.
- EU's DMA requires them to allow users to install third-party AI backends.
- Apple doesn't think parties other than themselves should be trusted with those iOS permissions.
I guess it'd be like if Apple allowed a first-party screen reader for iOS, so they refused to allow third-party screen readers.
Run by Apple where? Do they really have enough hardware to run it in-house?
See also https://security.apple.com/blog/private-cloud-compute/
It's not extreme interpretation, it's the intent.
Just say it would break your vendor lock-in.
EU has great intentions, and of course, feature parity should be offered so that competition can exist, but I don't find it crazy that it is more complicated on a product like that. As tech people things are very obvious to us but we need to remember that we are talking about a product used by everyone.
As an Apple user I feel more patronized than empowered here.
The AI provider would still be YOUR choice. You could stick with Apple's if you don't trust the other ones.
But why can Tesla ship Grok to their cars in the EU without any problems? Why aren't they required to let me choose between Grok, OpenAI etc or even a custom endpoint?
Simply because they are too small in user count. EU DMA, DSA etc. only apply at certain thresholds. Twitter for example falls under the scope, but Tesla is a distinct entity from Twitter and even if they were merged together, they would still be distinct services in the eye of the law.
It's the user's data. Not Apple's. And it should be the user's right to send it to whoever for whatever results, imo
For example, if I’m maintaining a secure chat app, I think I’d be more likely to adopt the APIs to share the chat messages with the system AI due to Apple’s promises that the data will either be processed On Device, or in their Private Compute Cloud.
If I instead believe that sharing the chat messages with the system AI would cause those messages to be sent to unknown-to-me other entities, I think I’d be less likely to participate in the new API.
This user might be okay with their data going to this other provider, but what about the people they’re messaging? I have a responsibility and a commitment to _all_ of my users to protect their data.
I might not be able to control what any specific user does with the data, but proactively writing the code that sends the chat messages to this other system is something that I have control over.
It's clearly just Apple not wanting to further open up their platform to competition.
Yeah, that's the whole fucking point.
Since it's the user's device, not Apple's, EU correctly "interprets" this as the user has the right to do whatever they please, including installing third-party chat apps.
Apple are just bulshitters when it comes to actual users, and not their corporate definition of a user.
BTW, did you know that in Japan, and in Japan only, you can change the Siri shortcut button to start other voice assistants? https://mjtsai.com/blog/2025/11/18/ios-26-2-third-party-voic...
Or that they wouldn't let you set default maps app outside of the EU: https://mjtsai.com/blog/2025/03/14/dma-compliance-default-ma...
It’s paternalistic, but I agree with Apple that free for all access to this kind of data is not a great idea. Ironically, before this could work we’d actually need much more EU style data regulation, and more consistently enforced.
Apple uses "privacy and security" as a cudgel to prevent anyone from breaking into the vendor lock in. To the point that EU actually had to explicitly tell Apple what to do [1], as Apple delayed features, made them extremely hard or convoluted for third-parties to use, and pulled every trick out of the malicious compliance manual.
This whole virtual assistants thing will drag on for another several years.
[1] https://digital-markets-act.ec.europa.eu/developer-portal/in...
Unless Apple proves otherwise I'm more inclined to believe they're either 1. Using this to try and shape the DMA in their own interest (definitely not their users' interest) or 2. Doing something with the data that would not be allowed in the EU (also not in their users' interest at all) or both.
https://www.macworld.com/article/3156959/apple-to-use-google...
People have to stop thinking Apple is somehow different.
> Apple’s going to try to run as much of the new Siri as possible on-device
Anthropic and OpenAI don't have edge models.
https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-de...
I primarily want Apple to provide extension points so that I can select my own provider, just as I can choose where to host my mail or install another app as my instant messenger.
Sure, I could install another provider's app, but it wouldn't have the same integrations, similar to how an instant messaging app would be less useful if notifications were limited to iMessage.
Apple doesn’t want to configure its private cloud to run every model. This seems fine.
The alternative would be to just stop invasive tracking and add the cookie when it’s actually needed.
For what it's worth, Apple claimed they proposed an "equivalent access" framework with some kind of "trusted agent framework" approach, but that it was shot down by the EU. I suspect it was way more inconvenient for third-party developers than Apple lets on.
“Make your reminder app’s actions available to Apple Intelligence and Siri by adopting schemas for common reminder actions.”
https://developer.apple.com/documentation/appintents/app-sch...
https://developer.apple.com/videos/play/wwdc2026/240
Yes, the "Apple needs to look at your data to do this, but we don't have any way to look at the data if we wanted to". That's impossible, unless they open souce iOS and let people take control over their devices, and let people self host inference, so people can check that there is no network traffic. If it is as they say, they could let people host it without any downsides.
As the consumer, this just sucks because it means no matter which phone platform you choose, you're getting the same thing underneath, and there's no way to avoid it (besides not using an assistant entirely, which I recognize a lot of people do, myself included).
The source also says > The new architecture centers on Apple Foundation Models co-developed with Google, which Apple says are adapted to run both on-device and on servers through its existing Private Cloud Compute infrastructure
Which could mean Google and Apple have trained some custom models, probably the on-device ones, specifically tailored towards Apple's hardware.
Not even Apple has access to it, by design.
Last year the announced they were working with OpenAI. It looks like this went nowhere, so it's not really surprising to see them try someone else.
Take it with a grain of salt but I don't think it's other AI providers that Apple is upset about. The DMA would require users to be able install any openclaw like thing onto their device with access to everything that Siri can access today. There are all sorts of arguments to be made here but I can understand why Apple feels this way and wants to offer a good experience here.
[1]: https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-de...
They don't claim that. All they said is "later in the EU as we look into privacy and security" after spending two hours saying how private and secure everything is.
DMA would force them to allow usage from other apps than their own and other assitants than Siri, especially for on-device models.
Edit I stand somewhat corrected but it's regular Apple bullshit: https://news.ycombinator.com/item?id=48451012
It didn't work out well for Yahoo.com. It turned out that Google Search was the value and yahoo.com just skin around it. It might be the same for Apple. Gemini is the valuable part, what particular device you use it on matters less.
As i understand it, no LLM is miles ahead of the others right now, especially when it comes to simple agentic stuff. Hell, Qwen3.6-35B-A3 quantized to 3bits running on an 8 year old consumer GPU handles most agentic stuff fine, if a bit slow.
Differences in LLMs boil down to mostly the harness and the compute to run the models. Even for high complexity tasks like coding, the differences between openai, anthropic, google, and the bigger qwen models aren’t that dramatic.
(Ok so $110b is all services revenue not just icloud, but icloud’s a solid chunk of that)
And to your point, Google has a massive balance sheet, produces their own AI chips, and is not going anywhere anytime soon.
Is it really all that difference from Apple defaulting to Google's search engine?
How exactly are they disadvantaging themselves? Perhaps expand on that opinion a bit, and the data/assumptions you're making in forming it.
My naive assumption is that they're going to do what everyone is doing: make tooling that lets you swap in any model.
I don't think it's fair to assume incompetence, on their part. I think it's much safer to assume they're doing what's best for them, and it's very clear to them what's best (money, support, etc).
What's the difference now? I would guess 9/10 people here would have a very hard time telling the models apart in a blind taste test.
Well, which ones are on my Mac locally?
Which ones are in my iPhone locally?
Apple was not going to hand over the keys to AI to just anyone.
Apple is a Fortune 5 company with a brand value alone worth more than any of these AI labs besides Google.
There's too much at stake for them to not play it safe. There's almost nothing to gain taking a risk
I don't see the same thing here. Google isn't making any money from being the assistant in Apple, so why would they pay to be it?
I guarantee you Google will start letting people pay to influence the output of the Gemini models once they figure out how to do it.
smart is a weird term, gemma4 is an amazing omni model better than qwen3.6 for non coding tasks (as for all Gemini models). For Apple Intelligence gemma4 makes a lot more sense.
Also openai and Jonny Ive (love from) are cooking some device — may be personal
Also important to remember how immature OpenAI and Anthropic are as companies. It would be a huge technical, legal, and reputational risk to commit to using them.
OpenAI/Anthropic have nothing in this segment.
https://www.apple.com/apple-intelligence/
You can even see difference in agent harnesses using the same model in the same company if you compare Gemini CLI with AntiGrav. They are different experiences.
I’m pretty sure Apple’s agent harness will be drastically different from Google’s even with the same model
Is there a meaningful distinction between the Gemini-powered models and Apple Foundation Models? Does that distinction vary for on-device vs hosted models? Are some models running on Apple's Private Cloud Compute and others running on Google iron?
Edit: they elaborated significantly in a "keynote tech-talk": [0]
According to Apple, there are five models:
On-Device
- AFM Core: Dense architecture; the standard next-gen on-device model
- AFM Core Advanced: Sparse architecture, natively multimodal; enables features like image understanding and expressive voices
Private Cloud Compute
- AFM Cloud: Workhorse server model optimized for latency and cost
- AFM Cloud Image: Image generation and editing
- AFM Cloud Pro: Most capable model, Gemini frontier-level quality, for complex reasoning and agentic tasks; runs on NVIDIA GPUs in Google's cloud under Apple's PCC privacy guarantees
Everything excluding Cloud Pro are custom models running on Apple Silicon, "refined" using Google Gemini. About Cloud Pro, they say "this is our most capable model with quality similar to Gemini frontier models." So I might read between the lines and say this is a wrapped Gemini.
It's a 3B Apple Foundation model.
https://machinelearning.apple.com/research/introducing-apple...
If you've got a mac, you can use this to play around with it:
https://apfel.franzai.com/
Google also awhile back announced being able to run full Gemini by leasing / renting hardware in your own datacenters so companies can train or access data without needing to send things to their datacenters. Nvidia based. Guessing Private Compute might just be Apple leasing a ton of those?
But beside that, I feel like the app variant got worse the day they've had that wwdc-style release thing recently.
Previously it was a sparring partner that could actually keep up. But now it just doesn't.
Truly a shame. And nothing that could be fixed by local models any time soon, given that you need the size for the (cross-)domain knowledge.
Search would be better without the added AI hallucinations above it. If I want an AI answer I'll go and ask Claude, the quality difference is huge.
That's not Gemini, that's AI Mode (in Search), they're different products built by fairly different part of Google (actually one is built by Deepmind).
(I don't think it's much comparable to https://gemini.google.com/app at least in the past you'd get very different results)
This is one of the most cash rich companies in the world and it has failed to have any position in the most critical technology development perhaps ever.
It's a clear signal that Apple became the most incredible operational/execution company under Tim Cook, but lost its innovation leadership.
To my understanding, these "outside experts" have to go through a vetting process by Apple first. There are no publicly available audits of the infrastructure planned right now.
After Google saw the iPhone (before the public), they pivoted their vision for Android (it was originally on blackberry-style hardware), and that's when the "thermonuclear war" started. Kind of interesting Steve Jobs would have showed the iPhone off like that, when something very similar happened with Bill Gates, which prompted the development of Windows (according to Jobs).
19 years later, it's probably time to be pragmatic again. If Apple isn't able to deliver on some of these AI integrations into the OS, they risk losing users to Android. If they have to pay someone for a model, they might as well choose the one they think is going to be best for their users. This keeps existing iPhone users on iPhone, and may pull over some Android users looking for the same features, but with better privacy. That seems like a win for Apple. To pay OpenAI instead of Google would just be spite at this point. Maybe well deserved, but the leadership has all changed over in the last two decades, so they'd be fighting old wars. Though I think they should still think twice before showing Google anything that hasn't yet been released.
Google probably gave them the best deal. When you're the #3 player you'll sacrifice margins to drive volume.
Gemini models clearly gaslight the user and hallucinate, they're also SUPER verbose, as shown in the demos from the keynote.
Plus, if they're not charging a subscription for this, you know we're getting the dumbest models...
Meaning the system prompt(s), harness, entry and exit points, and skills. So the product is still "Siri AI", because of all the stuff that takes it from a raw infrastructure concern upon up into a "product" is Apple's responsibility.
Google are "okay with that" because Apple pays them $1B a year, per press reports, to be.
What a blunder, they resisted AI for like 2 years when it was all the buzz, and now when the bubble is about to bust and every user has AI fatigue they decide to finally dip into the fad?
Before it was as if avoiding AI was a conscious design decision, and if there was an AI crash, Apple would be the only survivor left. Now it feels like they weren't in on the meme out of incompetence and are now late to the party.
No one can know what Jobs' stance would be, but I like to think he would be anti-slop
You can run smaller models on cheap commodity hardware.
Your phone can probably run one of these:
https://play.google.com/store/apps/details?id=com.google.ai....
https://apps.apple.com/us/app/google-ai-edge-gallery/id67496...
I’m not saying people who hold this view are being dishonest at all. But sometimes, to me, brands like Harley-Davidson or Apple seem closer to a cult than to a typical corporation.
I probably wouldn't use it without that. It's one thing sending my shitty code to be trained on, and another thing entirely to give these companies access to my personal life and information.
I suspect if you paid apple enough money, and were willing to prove that your personal Private Cloud Compute did meet their requirements, it wouldn’t be impossible.
The hardware isn't a real justification, just a convenient fig leaf.
When you have to image a highly irrational reason to explain why groups of people do the things they do, there's a decent chance you just don't understand their perspective. They may be acting reasonably rationally from their own perspective. (As you said yourself: "I honestly don't understand...")
Also I have seen that Apple has some strange lust towards image generation as if that's what people really want. I have this slop image generation thing on my phone and it is useless.
Here's what I want: natural language interaction to achieve complex workflows in iPhone. Example: find the cheapest way to go from A to B and book it using the Deutsche Bahn Train app.
If they don’t like this in the future they can just change to the less convenient, less secure, and likely more expensive bedrock + SOTA.
Even if it's a bad take, call out what's wrong with the take, rather than attack the author.