There has been a trend recently in computation proof of work only CAPTCHAs, e.g. mCAPTCHA, or Friendly CAPTCHA (https://proton.me/blog/captchas https://proton.m

There has been a trend recently in computation proof of work only CAPTCHAs, e.g. mCAPTCHA, or Friendly CAPTCHA (https://proton.me/blog/captchas https://proton.me/blog/captchas), but our own data shows that these are not sufficient to stop attackers due to compromises made to make PoW work for real users with slow devices. Adding the visual challenge has been essential in stopping these attackers. The combination of computational proof of work and human proof of work does in our experience provide defense in depth. However it's not perfect. One of the differentiators of ProtonCAPTCHA is that we've built this system with the expectation that someone will break it. So, as you've alluded to, in the event that someone or some thing is able to navigate these challenges either through automated mechanisms, or via third party solvers, we have defenses against such attacks/automations. That is a third hidden level of defense -> however, you will understand that for obvious reasons we do not divulge how this is done.