A patent is a wishlist, not a product. But when a company’s filings sit awkwardly next to its public message, the filing is usually the more honest document.
Every big company has two voices. One is the keynote: warm, careful, focused on you. The other is the patent filing: written by lawyers, read by almost no one, and far less concerned with how it sounds. We read the second voice all day, and one habit has stuck with us. When a company spends a lot of words in public on “privacy” or “safety” or “trust,” it is worth checking what those same words are doing inside its filings. Often they are sitting right next to “ad attribution,” “biometric capture,” or “monitoring.”
A fair warning before we start: a patent application is not a product, and it is not a confession. Companies file defensively, file things they will never build, and file to keep options open. None of what follows is proof that anyone lied. It is just the gap between the version of a company that goes on stage and the version that talks to the patent office. The gap is where the interesting reading is.
Google: “privacy-preserving” was always modifying “ads”
For years Google’s public story was the Privacy Sandbox, a set of technologies pitched as the privacy-friendly future of the web, the thing that would let Chrome kill the third-party cookie without breaking the ad-funded internet. The filings tracked that story closely. Google patented a privacy-preserving ad attribution system using attestation tokens and a sandboxed ad auction built on trusted execution environments.
Read those titles slowly. The adjective is “privacy-preserving.” The noun is “ad attribution.” The point was never to stop tracking you for ads. It was to find a way to keep doing it that survived regulators and headlines.
And here is the punchline the keynote never delivered: in October 2025, Google retired most of the Privacy Sandbox, and third-party cookies stayed in Chrome. The initiative that justified years of “privacy-first” messaging was quietly shelved. The patents remain, a fossil record of a privacy promise the company stopped keeping.
Microsoft: the off-switch is the admission
Microsoft’s public posture is trust. Your data is your data, the enterprise messaging goes, and we are the responsible adults of big tech. Then in 2024 it announced Recall, a Windows feature that took a screenshot of your screen every few seconds and stored it in a searchable database. The backlash was immediate and loud enough that Microsoft paused it, re-encrypted it, and made it opt-in.
The filings sit in exactly that tension. Microsoft patented a built-in Windows toggle for controlling AI data collection, and a prompt-security system that profiles both the AI models and the users interacting with them. An elegant off-switch for data collection is a useful thing to build. It is also an admission of what is being collected in the first place. You do not patent a tap unless the water is already running.
Meta: a privacy-focused vision, and a radar in the headset
In March 2019, Mark Zuckerberg published “A Privacy-Focused Vision for Social Networking,” promising encrypted, intimate, ephemeral spaces. It was the company’s big pivot after a rough few years.
The hardware filings are a different kind of intimate. Meta patented radar-based facial-expression tracking for VR headsets, a system that reads the movements of your face from the inside of the device you are wearing. To Meta’s credit, it has also filed for a face-verification system designed to never see your raw biometric data, which is a genuinely privacy-protective idea. Both instincts live in the same patent portfolio. The question the press release does not answer is which one wins when they conflict.
Apple: the case where the filing backs the promise
It would be easy, and unfair, to put every company in the same bucket. So here is the counterexample. Apple has built an entire brand on the line “Privacy. That’s iPhone,” and the Vision Pro gave it the most invasive sensor suite it has ever shipped: continuous eye tracking, hand tracking, and a 3D map of your face. We have read a stack of Apple filings on exactly this, including real-time eye tracking that updates a 3D model of your eye and gaze detection that triggers actions when you look at a corner of the screen.
Eye-movement data is some of the most revealing data a device can collect. It can hint at health conditions, attention, even what you want before you act. And here the public claim holds up: Apple processes that eye data on the device and does not hand the raw stream to apps, to websites, or to itself. The architecture in the filings matches the promise on the billboard. That is what it looks like when the two voices agree, and it is why the cases where they do not are worth flagging.
How to read the gap yourself
None of this requires assuming bad faith. The pattern is simpler and more useful than that. A keynote is an argument about how a company wants to be seen. A patent is a description of what it is actually trying to build, stripped of the adjectives that test well in focus groups.
So when you see “privacy-preserving” in a filing, finish the sentence: privacy-preserving what? When a company patents a control, ask what it is controlling. When the public message is loudest, the filings are often where the quiet part is written down. We will keep reading them, and we will keep noting when the two voices stop matching.