Apple's AI boss Craig Federighi has a message for OpenAI, Anthropic, and its other AI rivals: 'They're pursuing AI for the sake of…

Craig Federighi didn't mince words at WWDC 2026. Apple's software chief used one of the company's biggest stages of the year to call out the AI industry—not by name, but with enough specificity that nobody was left guessing. "Some appear to be racing forward, seemingly pursuing AI for the sake of AI, without clear regard for the people it's ultimately meant to serve," he said. The target was clear: the OpenAIs and Anthropics of the world. But his sharper point was about privacy. Most AI providers, he argued, make users fight for it—through temporary chats, manual deletion, or turning off features entirely. "At Apple, we believe privacy in AI is non-negotiable."It's a strong claim. And it's one that's easier to make when the alternative is looking increasingly precarious.A New York Times investigation earlier this year found that chatbot conversations are regularly surfacing in court proceedings, reviewed by company staff, and handed over to federal agents under legal requests. Anthropic has confirmed it shares data when legally required. OpenAI acknowledged reviewing a user's conversations months before she carried out a mass shooting in British Columbia—a case now reshaping debate around what AI companies are legally obligated to do with what they know. The intimacy of the problem is what makes it distinct from older privacy concerns. Unlike a web search, a conversation with an AI reveals intent, emotion, and context that a few keywords never could.This is the backdrop against which Apple is making its privacy pitch—and at WWDC, it went further than before.Private Cloud Compute, Apple's server-side architecture where user data is never stored or made accessible to anyone, is expanding for the first time to a third-party data centre—Google Cloud, running on NVIDIA GPUs. The core guarantees stay intact: stateless computation, no privileged runtime access, and verifiable transparency that outside security researchers can independently audit. All PCC binaries will be published for public inspection, with live research access available through Apple's Security Bounty Program. Google built something comparable last year—Private AI Compute, running on its own Gemini models and custom TPUs—with similar no-access assurances. The distinction Apple is pressing is end-to-end verifiability: not just a promise, but a system that can be checked, regardless of where the servers sit.For users, the practical upshot is straightforward. You shouldn't have to read a privacy policy, toggle a setting, or use incognito mode to keep your AI conversations from being stored, reviewed, or handed to a court. That, at least, is what Apple is arguing. Whether it delivers is what the next few years will show.