Apple built Private Cloud Compute — AI servers that process your data and delete it. They publish the code for verification. This is genuinely better than what Google or Microsoft do. Then they added ChatGPT. When Siri can't answer, it asks: "Want me to send this to ChatGPT?" Your query leaves Apple's infrastructure and goes to OpenAI — a company 27% owned by Microsoft, now also running on Amazon's cloud. Apple asks permission first. But the default is there. The nudge exists. And every time you say yes, your data leaves the ecosystem Apple spent years building walls around. Apple Intelligence is only on iPhone 15 Pro and newer. If you have an older iPhone, you don't get on-device AI — you get pushed toward ChatGPT and Google, which have weaker privacy. Privacy is now a hardware paywall. And the AI that does run on-device hallucinates. Apple's notification summaries invented fake news headlines and attributed them to the BBC. The BBC complained publicly. Apple's privacy-first AI fabricated journalism and put a real news organisation's name on it. On-device processing doesn't matter if the output is false. Private lies are still lies.
What they claim: Apple's privacy labels and App Tracking Transparency (ATT) position iPhone as protecting users from third-party tracking.
What we found: While ATT reduced third-party tracking through apps, Apple simultaneously grew its own advertising business. Apple Search Ads revenue grew to an estimated $7-10 billion annually by 2025. Apple's own apps are exempt from the ATT prompt — Apple can track you across its own services (App Store, News, Stocks, Maps) without triggering the "Ask App Not to Track" dialog. Apple restricted competitors' ability to track users while expanding its own first-party data collection for advertising. The privacy feature that protects you from Facebook also protects Apple's ad revenue from Facebook. Privacy is real. The motive is mixed.
What they claim: Apple markets Apple Intelligence as privacy-preserving AI that processes on-device or through "Private Cloud Compute."
What we found: Apple Intelligence routes complex queries to Private Cloud Compute (PCC) — Apple Silicon servers running in Apple data centres. Apple publishes the PCC software for independent verification and claims no data is retained after processing. This is a genuine architectural advantage over competitors. However, Apple Intelligence also integrates ChatGPT (OpenAI) for queries it can't handle. When Siri escalates to ChatGPT, your query crosses from Apple's infrastructure to OpenAI's — a different company with different privacy practices, under different legal jurisdiction exposure (OpenAI now deploys on AWS as well as Azure). Apple asks permission before sending to ChatGPT, but the integration normalises cross-company AI processing on the device most people trust.
What they claim: Apple positions iPhone as the privacy-first smartphone: "What happens on your iPhone stays on your iPhone."
What we found: Apple Intelligence requires an iPhone 15 Pro or later — older iPhones are excluded from AI features. This creates a privacy tier system: users who can afford the latest hardware get on-device AI processing. Users on older devices are pushed toward cloud-based alternatives (ChatGPT, Google) that have weaker privacy protections. Apple's notification summaries — an Apple Intelligence feature — have generated embarrassing errors, including false news headlines attributed to real sources like the BBC. The BBC complained publicly. Apple's AI misattributed fabricated news to the BBC's brand. On-device privacy is meaningless if the AI hallucinates false information and attributes it to trusted sources.