Re: HTTP HashCash

On Thu, May 22, 2025 at 6:25 PM Demi Marie Obenour
<demiobenour@gmail.com> wrote:
>
> On 5/22/25 13:37, Watson Ladd wrote:
> > On Thu, May 22, 2025 at 1:15 PM Demi Marie Obenour
> > <demiobenour@gmail.com> wrote:
> >>
> >> On 5/22/25 08:59, Ben Schwartz wrote:
> >>> In general, the IETF has been skeptical of "proof of work" designs that deliberately waste CPU time.  As an alternative, you may want to review Privacy Pass (RFC 9576-9578), which allows an HTTP Origin to require clients to expend a different kind of resource ("tokens") that may be limited, without learning the clients' identities.
> >>
> >> Does that just move the problem to the token issuer?
> >
> > And from the shameless plug department, that is why privacypass exists!
> >
> > Token issuers can have much better ways to issue limited use tokens:
> > they may be aware of hardware support on the client to limit identify
> > proliferation, or existing relationships that make bypassing
> > expensive. This capabilities cannot usually be expressed over the
> > Internet without significant privacy impacts (but read
> > https://www.usenix.org/conference/soups2022/presentation/whalen for an
> > alternative, and the accompanying SAC 21 paper to see how the crypto
> > is done (in a way that's rapidly deployable: production at Internet
> > scale with browser support would make different tradeoffs)).
> My concern is that these methods are going to be used to deny service to
> those using non-attestable open systems such as those running desktop Linux,
> or to systems running alternate operating systems such as GrapheneOS.  For
> users to be denied access to a website because of this, or to be forced to
> upload a government-issued ID (which they might not have), would be very,
> *very* bad.  Proof of work is indeed incredibly inefficient, but it does
> not have these risks, as it can be passed by *any* device with enough
> time or processing power.

Hardware attestation is only one potential signal: account
authenticity signals from long running accounts (think email/social
media) can also be used.

The relative power of attacker scraping or engaging in other kinds of
exploitation to legitimate users can be much larger than you might
imagine if the rewards are high enough. Existing techniques in this
field are pretty terrible for ordinary users on unusual devices or
browsers.
> --
> Sincerely,
> Demi Marie Obenour (she/her/hers)


--
Astra mortemque praestare gradatim

Received on Friday, 30 May 2025 21:06:54 UTC