Turning off tracking isn't enough: the Google verdict lesson

In September 2025, a U.S. federal jury ordered Google to pay $425.7 million: for years, the company kept collecting the activity of roughly 98 million users who had turned off the very setting meant to stop it (“Web & App Activity”), via third-party apps. Google is appealing. The lesson goes beyond Google: a privacy toggle is a promise, not a guarantee. As long as you hand over the data, you depend on the provider to honor your settings. The only guarantee lies in what you don't transmit.

What the court found

In Rodriguez v. Google, the jury found an invasion of privacy: according to reporting by Reuters and CNBC, Google kept gathering data from third-party apps after users disabled the setting presented as the way to stop that collection. The figure — $425.7M, a few dollars per person across a class of about 98 million — matters less than the principle established: the gap between what a setting promises and what actually happens can run for years before it's sanctioned.

A setting is a promise, not a guarantee

Turning off a tracking option doesn't physically remove the data: it asks a remote system not to record it. Between your click and the actual behavior sit code, third-party partners, legal interpretations and mistakes. The outcome depends entirely on correct execution on the provider's side — exactly what the verdict called into question.

The same principle with AI

Pasting text into an AI rests on the same trust. The “don't train on my data,” “history off” or “temporary chat” options are useful — but they're provider-side settings you can neither inspect nor prove. If one is misapplied, misunderstood or changed, it's your real data on the line. And once transmitted, you can't recall it.

What you don't control, once the data is sent:

• whether the setting is actually honored, and on which channels;
• what downstream subprocessors and partners do;
• policy changes, technical logs and legal obligations.

What you actually control

Only one thing doesn't depend on a promise: what you don't transmit. If the sensitive data never leaves your device, no misapplied setting can expose it.

1. 1Don't rely on a remote switch to protect sensitive data.
2. 2Remove the sensitive data before sending, instead of hoping it's ignored.
3. 3Keep the token↔value mapping local, out of the provider's reach.

That's ONYRI Sanitize's logic: the engine detects sensitive data and replaces it with reversible tokens before sending; detection and the token↔value mapping stay in your browser, and only anonymized text reaches the AI. You no longer depend on a provider-side setting — the real data never left in the first place.