[cobertos]

That's not how he's suggesting it would work. All checks would normally be applied to build a "how human are you" or "humanness" score. He's suggesting a pwned email test and arguing it would be a good signal for "humanness". The implementation might not make it an explicit penalty (-1 to your "humanness" score), but not being pwned might not help your case (+1 if you are pwned, but +0 if you're not).

[TheCleric]

Yeah it would definitely be good to integrate it into a Bayesian approach where it is mixed with other factors to generate a % chance of being human vs. bot.