Every guide, comparison and detector run we have published under fingerprinting.
Any page can make Chrome download gigabytes of on-device AI models, once per profile, through that profile's proxy. The two obvious switches either change what the page sees or keep the download and stop security updates, so the fix went into the engine.
A page draws text to a hidden canvas, reads the pixels back and hashes them. The result is stable on one machine and different across machines, which is exactly what makes it useful for tracking and hard to fake convincingly.
Chrome 153 reached the stable channel on 2026-08-26 and the AntiBrow kernel now builds on it. The number in your user agent is the least interesting part of that. Here is what a detector reads instead, and why moving every profile to the newest build on release day is its own tell.
Detection stopped asking which fonts you have and started measuring how wide they are. A missing font is not hidden: it renders in the fallback and measures exactly like a family nobody has ever installed. If your persona claims Windows on a machine that is not Windows, that is a one-line contradiction, and no amount of aged profiles or clean residential exits talks you out of it.
A high anonymity score means little. What gets profiles flagged is contradiction between signals, not any single value. A checklist of the pairs that must agree, and how to test each one in a few minutes.
A page asks your GPU to identify itself, then measures dozens of rendering limits and a drawn scene's pixels. No permission prompt is involved, and the result is one of the most informative signals a browser exposes.
Fingerprinting identifies you from the combination of hundreds of ordinary browser properties, no cookies required. The counterintuitive part: aggressive blocking usually makes you more identifiable, not less.