The current landscape of CAPTCHA technology is pretty bleak. It's pretty easy to use ML to learn and solve the early first-gen CAPTCHAs that just used crossed-out words. Google reCAPTCHA relies primarily on user data, obfuscation, and browser fingerprinting to filter out bots, but that only works because of (possibly misplaced) trust in Google. It falls back to an image recognition challenge (which hCaptcha uses exclusively) if you don't have a good data profile - which can also be solved by automated means.
I don't see desktop Linux being fully untrusted off the Internet, if only because Google won't let it happen. They banned Windows workstations internally over a decade ago and they are institutionally reliant upon Linux and macOS. What will almost certainly happen is that Linux will be relegated to forwarding attestation responses between Pluton, some annoying blob in Google Chrome, and any web service that does not want to be flooded with bots in our new hellscape of post-scarcity automation.