Apple's New AI Models Are Built With Gemini but Designed for Privacy

Apple is turning to a longtime rival for artificial intelligence firepower, but with a twist that could redefine how big tech handles user data. The company known for locking down privacy is now building its Apple Intelligence features on Google's Gemini models — a move that signals both the intensity of the AI arms race and a potential new standard for data protection.

At its Worldwide Developers Conference 2026, Apple announced a partnership with Google that will see Gemini powering key capabilities inside Apple Intelligence, the company's suite of on-device and cloud AI tools. The deal covers image generation, text summarisation, and advanced natural language processing, giving Siri and other Apple services a major upgrade without compromising the company's long-standing privacy promises.

The partnership marks a stunning reversal from Apple's usual do-it-yourself approach to core technologies. For years, Apple has built its own chips, operating systems, and services to control both performance and security. But the generative AI explosion — dominated by OpenAI, Google, and Microsoft — has forced Tim Cook's team to rethink. Rather than building a foundational model from scratch, Apple is buying in, and from its biggest competitor in mobile operating systems.

Under the deal, Apple will run Gemini models in two modes: on-device for tasks like real-time photo editing and reply suggestions, and in a new 'Private Cloud Compute' environment for heavy lifting such as multi-step planning or document analysis. Apple says no user data is stored or accessible by Google, and all queries are anonymised and encrypted before reaching Gemini's inference servers. The system uses 'differential privacy' techniques and a verifiable audit trail, Apple claims, making it 'the most privacy-preserving large-language model integration ever deployed at scale.'

Apple has not disclosed financial terms, but sources close to the negotiation say the multiyear agreement is worth several billion dollars and includes revenue-sharing on AI-generated services like custom emoji and photo filters. Google, for its part, gets a massive distribution channel: every iPhone and iPad sold after 2026 will ship with Gemini-enhanced Apple Intelligence enabled by default.

The move has drawn sharp reactions. Privacy advocates worry that any cloud-based AI, even with Apple's safeguards, creates a new attack surface. But many security researchers praised Apple's approach, noting that its custom silicon includes a dedicated 'Neural Security Enclave' that isolates AI workloads from the main operating system. 'Apple is doing something genuinely new here,' said Dr. Elena Vos of Stanford's Center for AI Safety. 'They're treating privacy not as a feature but as an architectural constraint — and forcing Google to meet their bar.'

Looking ahead, Apple plans to open Apple Intelligence to third-party developers via a private API later this year, allowing apps to tap into Gemini-powered features while still respecting user opt-ins. The company is also rumoured to be building its own smaller language model for future devices, suggesting the Gemini partnership may be a bridge to eventual independence. For now, Apple has placed a high-stakes bet: that its users will embrace AI — as long as it comes with the privacy promises they've come to expect.