When Apple previews iOS 27 at its WWDC keynote on June 8, the most significant thing on stage will not be a device. It will be a three-tier rethink of how Siri works.
The company that spent years keeping Siri on-device is now sending its most complex queries to the cloud. Specifically, to a custom version of Google Gemini that Apple is reportedly paying roughly $1 billion per year to use. That is not a small bet for a company that built its brand on privacy-first computing.
The shift becomes concrete in September when iOS 27 lands. Users will notice it in two distinct ways. First, Siri gets meaningfully smarter. Summarizing your last ten emails, planning a multi-city trip, having a conversation that actually remembers context from earlier in the session, those capabilities arrive on your iPhone without you needing to open a browser. Second, the Photos app gets a dedicated Apple Intelligence Tools section with three new features that will make Clean Up look like a warm-up act.
Here is what is actually arriving, what might slip, and what the Gemini deal means for your data.
The Siri Upgrade: What Changes and What Does Not
Thomas Kurian, Google Cloud's CEO, confirmed at the Cloud Next conference in April that Apple is now a strategic Google Cloud customer. Gemini-based Apple Foundation Models will power a more personalized Siri arriving later in 2026. The deal terms, reported by Bloomberg and widely accepted in the industry, involve a custom 1.2-trillion-parameter Gemini variant and a multi-year commitment worth approximately $1 billion annually to Google.
Before users panic about their voice requests going to Google's servers, Apple has been clear about the routing. Simple commands stay on-device. Setting a timer, turning off Wi-Fi, playing a specific album from last Tuesday, those run entirely on the Neural Engine inside your iPhone. Nothing leaves the device. This architecture is not changing.
What changes is the heavy lifting. When you ask Siri to summarize your last three emails from your manager, to plan a weekend in Lisbon and drop flights into your calendar, or to look at a photo and tell you what plant it is, those requests route to cloud-based Apple Foundation Models that sit on top of Google's Gemini infrastructure. The request goes to Google's cloud infrastructure, gets processed by Apple's custom model layer, and returns. That is the new pipeline.
Around 60 Siri engineers were sent to a multi-week AI coding bootcamp in April, according to reporting by The Information. The timing, two months before the expected WWDC reveal, signals internal urgency. Apple is not arriving late to the AI assistant race with a half-baked product.
The Photo Tools: Extend, Enhance, Reframe
iOS 27 adds three new options to the Photos app editing interface: Extend, Enhance, and Reframe. Each does something meaningfully different from what Clean Up, the only AI editing tool Apple currently offers, can manage.
Extend uses generative AI to extend the edges of a photo beyond what was originally captured. Change the crop to a wider aspect ratio, and the tool fills in the missing scenery rather than leaving blank space at the sides. The capability mirrors what Google introduced with its Magic Eraser and Best Take features, and what Samsung has offered in Galaxy AI for over a year. Apple is following, but the implementation reportedly includes gesture-based zoom controls that expand the frame interactively.
Enhance automates adjustments to color, lighting, and image parameters. Think of it as a smarter auto-correct for photographs, one that learns your preferences over time but does not require you to open a separate editing app. Apple has offered some of this functionality through the Photos editing interface for years, but the AI version promises more contextual adjustments.
Reframe is the most technically ambitious of the three. It works specifically with spatial photos captured on iPhone models that support spatial capture, and it lets users shift the perspective of an image after it has been taken. This is a feature with a narrow audience right now, but it signals where Apple sees computational photography heading as spatial capture becomes more mainstream.
All three tools could face delays. Apple has reportedly struggled to get Extend and Reframe working to its quality standard internally. Clean Up, which shipped in iOS 18.2, still had functional issues a year and a half after launch, according to MacRumors. That track record is worth noting. These features may not arrive on day one of the iOS 27 release, or they may ship in a reduced form.
The Privacy Question Nobody Is Asking Loudly
The Gemini-Siri integration raises a question that the tech press has not fully pressed: what happens to your Siri query data when it routes through Google's cloud?
Apple's stated architecture keeps simple commands on-device as before. For cloud-routed requests, the data flows to Apple's cloud infrastructure, which runs on Google Cloud. This is a different arrangement than asking ChatGPT or Gemini directly, where your prompts become training data. Apple's model sits in the middle, and the company has historically maintained that it does not use consumer queries for model training without explicit permission.
But the $1 billion question, quite literally, is what guarantees Apple has negotiated in its cloud contract with Google. The joint statement from January 2026 announcing the collaboration said Apple Foundation Models would be based on Gemini, but it did not specify data retention, processing location, or audit rights for Apple over Google's infrastructure.
Privacy-conscious users will want to wait for the fine print. The good news is that if you do not want cloud-routed Siri queries, the architecture suggests you will be able to disable that layer. The bad news is that the most powerful Siri features will not work without it. That is the trade-off Apple is asking users to make, and it is a more significant ask than the marketing language suggests.
The Lawsuit in the Background
Separate from the Siri overhaul, Apple faces a class action lawsuit filed in California federal court in early April. Three YouTube channel operators, Ted Entertainment, Matt Fisher, and Golfholics, allege that Apple circumvented YouTube's anti-scraping protections to download millions of videos. The complaint accuses Apple of using the content to train its AI models without permission from creators.
The lawsuit seeks maximum statutory damages under the DMCA, an injunction blocking Apple from continuing to use scraped material, and a declaration that Apple willfully circumvented YouTube's technical protections. This is not a fringe claim. The case has enough procedural weight that it will likely survive the initial motion to dismiss.
It creates an uncomfortable backdrop for a software launch centered on AI capabilities. Apple will spend the months leading up to WWDC talking about smarter, more capable AI across iOS 27. Meanwhile, its legal team will be defending the proposition that it did not train those AI systems on content pulled from the internet without permission. The two narratives will sit uncomfortably next to each other in coverage, and Apple's communications team will need to manage that tension carefully.
The Bottom Line
iOS 27 is Apple's most consequential AI software release since the Neural Engine shipped inside the A11 chip in 2017. The combination of cloud-routed Gemini intelligence, on-device Apple Intelligence photo tools, and expanded Visual Intelligence across system apps represents a coherent vision.
The realistic timeline is: WWDC preview June 8, developer beta immediately after, public release in September. The photo editing tools may arrive in reduced form or post-launch. The Gemini-powered Siri will probably ship in the initial release, given the engineering investment already confirmed by Google's CEO and Apple's internal bootcamp schedule.
The privacy trade-off is real but manageable for most users. The lawsuit is a separate risk that will not directly affect the software rollout but will generate press coverage that Apple would prefer not to have alongside an AI launch.
For iPhone owners who have been watching Android competitors ship AI features for two years, iOS 27 is the moment Apple stops being late and starts being evaluated on whether it did it better.
That evaluation begins in June.