Original comment follows:
In my view, this would just DRM-ize everything on the web. Of course, Cloudflare and Fastly don't talk about this much, and Cloudflare keeps assuring you'll still get captchas if device attestation fails or is unsupported. But realistically, once all Microsoft, Google and Apple implement it in their devices, there isn't much of a reason to keep accepting non-attested devices. You can already see where this is starting to go - if you're using Linux/BSD or another niche OS, congratulations, you can't submit forms any more. And since device verification would become extremely cheap to perform this way, you'd also see websites protected entirely by this tech, effectively locking out Linux/BSD users. The Cloudflare article also talks about how, at least in the case of Apple, they'd run something like a posture assessment to confirm that your device components are genuine. I can also see this new tech locking out users of non-OEM repairs. This is a much bigger deal than what it seems like on the surface, and I'm genuinely scared about how this one simple move dwarfs all of the "evil" things that big tech has done so far.
https://www.ietf.org/archive/id/draft-private-access-tokens-...
https://www.ietf.org/archive/id/draft-private-access-tokens-...
We shouldn't fight bots. We should use trust instead. Not global trust, it must be subjective. I trust A, B, C. B trusts D, E. E trusts F. It should be weighted. There's small world effect [2]. There's just a few hops between any two people in the world. It solves SPAM, it solves reviews, scam, news and maybe politics. Somebody please get it done already.
1. https://duckduckgo.com/?q=phone+farm+bots&t=ffab&iar=images&...
Besides it's an open standard. https://www.ietf.org/archive/id/draft-private-access-tokens-...
Since I wrote it I became confident that algorithm which is used for cumulative trust computation should be up to each node (instead of using zk-SNARks for example). If you trust somebody, you trust them to compute it as they wish. And I would drop dimensionality at least in the beginning. Probably using multiple identities in place of it.
The thing that strikes me is that they bring up Privacy Pass (https://privacypass.github.io/) as related work, and while I've never been completely, totally on board with Privacy Pass, I also feel like the reliance on hardware/OS verification checks here is strictly worse than what Privacy Pass is offering?
Forget the user experience for a second and privacy implications (Privacy Pass at least seems to be mostly hardware independent and can work on any device/browser that implements an extension, which has comparatively fewer negative implications for a competitive indie web ecosystem) -- speaking purely as a website operator, hardware checks seem strictly easier to game than a CAPTCHA. So even if I'm not a user trying to use a device that doesn't have these attestation schemes built into it, if I'm an operator wouldn't I prefer to have a protection that's harder to bypass by a click farm?
I'm not saying I would be completely thrilled with Privacy Pass either (CAPTCHAs in general are accessibility problems). But should I be thrilled about a version of Privacy Pass that (as far as I can tell) inherently must be more invasive to my hardware, and that isn't guaranteed to work on every device/browser that I use?
I mean the metadata kinda indicates otherwise: https://developer.apple.com/news/?id=huqjyh7k