Ryan Whitham:

The issue before the commission currently is the built-in advantage for Gemini on Android. When you turn on any Google-powered Android phone, Gemini is already there and gets special treatment at the system level. The European Commission is taking aim at the lack of features available to third-party AI services. The commission believes that there are too many experiences on Android that only work with Google’s Gemini AI, and as a gatekeeper, Google must change that.

[…]

European regulators are proposing several broad changes to the way AI tools operate on Android phones. Some of this is straightforward, like allowing third-party AI tools to be invoked system-wide via hot words or button presses. This might also include allowing AI tools to view screen context when the user opens them.

[…]

Many of the Gemini AI features in Android, including Magic Cue, rely on running local models, and Google has been slow to allow third parties the system access to make that work effectively. So the EU is also suggesting a mandate that would ensure developers have the necessary hardware access to run local models “with high levels of performance, availability and responsiveness.”

Via John Gruber:

The difference between Google and Apple on this front is that Google just blazed ahead and shipped Gemini integrated into Android in the EU, and is now facing compliance problems after shipping. (Ask forgiveness.) Apple isn’t shipping Siri AI in the EU in iOS 27, knowing that it’s going to be deemed non-compliant. (Ask permission.)

Previously:

• No Siri AI in EU
• DMA Compliance: Google
• Can Anyone But a Tech Giant Build the Next Big Thing?

Update (2026-07-17): Juli Clover:

The European Commission today formally ordered Google to give third-party AI services the same access to Android device features that Gemini has.

[…]

The European Commission is giving Google a full year to implement changes that will meet the Digital Markets Act requirements, and that's before any legal appeals Google might make.

Google (via Ben Schoon):

This Android ruling threatens device security by granting external apps sensitive and powerful device permissions without these safeguards.

Update (2026-07-23): John Gruber (Mastodon):

They provide separate “Q&A” overviews of the guidance for Android AI interoperability and web search sharing, and the full guidance documents are PDFs (Case DMA.100220 for Android AI, Case DMA.100209 for web search).

[…]

The EC is demanding that Google create APIs that allow third-party AI assistants to do everything Google Gemini does now[…]

[…]

Also, in my reading, the EC is demanding that Google make available to third-party AI assistants all information in Google’s own apps (Gmail, Google Calendar, Google Docs, Google Maps, etc.) that Gemini has access to.

[…]

Apple unfortunately hasn’t shared any technical details describing its proposal for a “Trusted System Agent” that it shared with the EC last year. But whatever Apple’s vision for the Trusted System Agent is, I don’t see how the EC would deem it compliant with the DMA if they want from Apple anything close to what they are now demanding from Google.

I don’t know what will come of this, but I’m pleased to see it in the sense that it is logically coherent. The TSA seems like a way to give the appearance of doing something without actually solving the problem. Like Gruber, I don’t really see why Google would want to comply with this rather than taking their Gemini ball and going home. Unlike Gruber, I don’t find it scary to let users opt into sharing their data and background time with apps of their choosing. I don’t like software design by government committee, but it seems less bad than further entrenching an unfettered monopoly. Maybe this could have been avoided if Google had developed its own reasonable solution instead of shipping something that was obviously not compliant.

Apple (xip, downloads):

Xcode 26.6 includes Swift 6.3 and SDKs for iOS 26.5, iPadOS 26.5, tvOS 26.5, macOS 26.5, and visionOS 26.5. Xcode 26.6 supports on-device debugging in iOS 15 and later, tvOS 15 and later, watchOS 8 and later, and visionOS. Xcode 26.6 requires a Mac running macOS Tahoe 26.2 or later.

[…]

Google Gemini is now available in the coding assistant.

Xcode adds support for the Agent Client protocol.

[…]

Launch Time similar-app goals have been refined for improved accuracy, establishing new baselines.

Previously:

• Xcode 27 Announced
• Xcode 26.5

Apple (MacRumors):

Apple today introduced the next generation of Apple Intelligence, powered by a bold new architecture that integrates the latest Apple Foundation Models deep into Apple’s platforms and is uniquely designed to protect users’ privacy.

[…]

The next generation of Apple Intelligence also helps power Siri AI, an entirely new version of Siri.

[…]

Image Playground offers new powerful ways for users to bring their imagination to life. They can create high-quality images in virtually any style, now including photorealistic, thanks to a new generative model that runs on Private Cloud Compute. This is a major transformation for image generation across platforms. And generated images will automatically include a hidden SynthID watermark to identify them as AI-generated.

[…]

Now Messages offers one-tap suggestions based on the context of users’ conversations, making it easier than ever to get things done, such as creating a reminder or a note.

[…]

Apple Intelligence powers even more enhancements across operating systems. With automatic proofreading, users receive improved suggestions for spelling and grammar as they type across the system.

Damien Petrilli:

Apple WWDC26 “big features”: selling a rebranded gemini.

John Gruber:

What’s confusing about this Apple-Google partnership is that Google pretty much calls all things AI “Gemini”. The models are “Gemini”, the assistant is “Gemini”, and the feature integrations are “Gemini”. So Apple is taking pains to emphasize that they’re building atop the Gemini models, not the Gemini assistant.

Google creating the model, Apple figuring out how to integrate it and make it easy to use. Are we back to Apple and Google each focusing on what they do best?

Dave B.:

This is my favorite Apple event in a long time.

Instead of bombarding you with a million new features that mostly feel half-assed, this feels like it’s actually catered to the user experience instead of to marketing checklists.

Matt Ronge:

All of these AI additions to MacOS/iOS look like nice improvements, but somehow it all feels unambitious given the current state of AI.

But I suppose it’s better for them to underpromise given what happened with Siri last year…

Christina Warren:

As I expected, Apple is going to punt the “on-device” story for Apple Intelligence and push towards the “private cloud compute” story for the models that you’ll actually want to use. I’m glad on-device isn’t going away, but it’s clear a hybrid approach is absolutely necessary

Hartley Charlton:

Apple’s most advanced on-device AI model in iOS 27 requires a minimum of 12GB of unified memory, meaning the standard iPhone 17 is excluded.

Alex Rosenberg:

The 16 Pro was explicitly sold as having Apple Intelligence-capable hardware.

Juli Clover:

Apple today said it is expanding Private Cloud Compute (PCC) beyond its data centers, partnering with Google and NVIDIA to run Apple Intelligence workloads on Google Cloud.

[…]

All server components and software are part of a trusted computing base subject to verifiable transparency and no-privileged-access guarantees, plus Apple has a cryptographically verifiable ledger of all Google Cloud hardware that is part of the PCC fleet to mitigate the risk of supply chain attacks. PCC on Google Cloud also uses many of the same architectural security patterns as PCC on Apple silicon.

John Siracusa:

PSA: Remember that Apple Intelligence is disabled in macOS if you boot from an external drive.

See also: Mac Power Users Talk.

Previously:

• Apple Foundation Models in appleOS 27
• Photos AI in appleOS 27
• Agentic Password Updates
• AI-Generated Shortcuts
• Safari 27 Announced
• Siri AI Announced
• WWDC 2026 Keynote
• Gemini-Powered Siri Features Delayed
• Apple Picks Gemini
• White Label Gemini on Private Cloud Compute

Update (2026-06-19): Apple:

As we continue to refine our approach to image generation, the ImageCreator class is being discontinued and will no longer work in iOS 27, iPadOS 27, macOS 27, and visionOS 27 or later.

[…]

Transition to presenting the Image Playground sheet, which provides a consistent, system-managed image generation experience. Alternatively, you can integrate another image generation service of your choice.

Apple:

At the heart of this architecture is our third generation of Apple Foundation Models (AFM), a family of five foundation models custom-built in collaboration with Google. These span from on-device models to server-based models running on Private Cloud Compute.

Apple Foundation Models are built to unlock a wide range of helpful experiences for our users, like an entirely new Siri and intelligent tools that make everyday apps smarter and more useful.

Hartley Charlton:

“The amount of the Google Assistant we use is none,” Federighi said, explaining that Apple 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.

Federico Viticci:

The cloud models hosted on Private Cloud Compute are interesting: they’re not “Gemini” models; they’re Apple Foundation Models trained using proprietary data via RL, which were then “refined” with data from Google’s “frontier” models.

[…]

I’ve been covering the AFM family of models for a while now, and AFM Core Advanced is one of the most interesting on-device models I’ve read about in a while, especially in the context of model size for mobile devices and built-in multimodality with support for text, images, and audio. I’m very keen to play around with this model and understand how it holds up in practice. I wonder if the new CLI (!) for AFM will let you test this one.

Awni Hannun:

It’s very cool that Apple shipped a 20B parameter on-device.

You can’t put 20B parameters in RAM at any reasonable precision. To make it work they are using pretty exotic architecture by today’s standards.

A small model predicts from the query (or prompt) which experts to load from Nand into RAM. The key distinction from a typical MoE is that you do this once per query and then generate all the tokens with the same experts (instead of switching the experts for every token).

Greg Pierce:

Looks like there’s still a 4K context size on local FoundationModels. Pretty limiting.

Kyle Howells:

Really disappointed SwiftUI’s result builder syntax has infected FoundationModels API too.

How did a million modifier methods become the default way to build APIs?

It’s horrible for discoverability, code completion, documentation.

Steve Troughton-Smith:

FoundationModels is one of the highlights of this WWDC; I think it’s now finally what we hoped we were getting two years ago, and thought we were getting last year (just the models themselves were barely up for it then). I finally feel like I can build features with it, though I really don’t like the Sword of Damocles Apple is holding above Private Cloud Compute and how it can permanently lock you out of using the models forever across all your apps if you hit a download threshold. Poison pill

John Gruber (Mastodon):

These strict limits don’t seem to be getting as much attention as they should. It’s nice that for small developers who meet the above criteria, access to PCC has no cost. But there’s no way (yet?) to buy your way out of these limits.

[…]

The “fewer than 2 million first-time app downloads from any of their apps” restriction is particularly notable. It’s not 2 million installations for apps that are using PCC, but 2 million downloads for any app the developer has ever released.

Chad Podoski:

Is there a real time count somewhere of total lifetime downloads for a developer account, so one can plan if they at risk of the switch tripping? Reminds me of the small business qualifications. I had to try to figure out if an employers developer account was going to exceed the revenue threshold for the year. I was left trying to determine it from sales reports, not to mention sales date versus pay out date confusion. I think I botched the calculation in the end.

Anthropic (Hacker News):

Claude for Foundation Models is a Swift package that makes Claude available as a server-side language model in Apple’s Foundation Models framework. The package conforms Claude to the framework’s LanguageModel protocol, so you drive it with the same LanguageModelSession API you use for Apple’s on-device model: respond(to:), streaming, guided generation, and tool calling all work the same way.

Requests go directly from your app to the Claude API; Apple is not in the request path and does not see prompts or responses. Usage is billed to your Anthropic account at standard API pricing.

Previously:

• Apple Intelligence in appleOS 27
• Siri AI Announced
• OmniFocus 4.8.10
• OmniOutliner 6.1
• Foundation Models Framework
• Apple Intelligence Foundation Language Models

Update (2026-06-17): Simon B. Støvring:

Apple is literally stopping me from building a brand new app using Private Cloud Compute because I made a successful app a few years ago.

The new app might be a success, or it might be a complete flop. Nobody knows yet.

I would understand it if Apple were to limit developers once an app reaches a certain number of downloads or active users, but that’s not what’s happening here.

Instead, they’re blocking developers before they’ve even had a chance to build and ship the app.