[peeters]

In the article they also say:

> Again, this is entirely rational for Google. If the value of the image classification training did not exceed those costs, it makes perfect sense for Google to ask for payment for the service they provide.

This might be exacerbated in the case of Cloudfare. Imagine a system where 99% of the visitors being challenged are human. The data gathered from such visitors is quiet, quality data. That fits the usecase of validating an anonymous poster on some random blog. Now consider the Cloudflare usecase. Visitors will only be challenged when Cloudflare already expects you're a bot. Most of the challenges are served to bots. The data is much lower quality, but their cost per challenge has remained the same.

It could just be that as this type of usecase became dominant, the balance of value tipped.

[gurrone]

I guess this is very true. Our quite elaborate Cloudflare Firewall setup combining bot management scores with GeoIP and network information to decide on the action has solve rates below 0.5% on most rules.

The only case where we see up to 3% solved is on rules targeting networks which contain mostly free (as in beer) VPN providers (the new pest of the internet). Those networks sent a lot of malicious and automated traffic with the mixed in 3% of real users.

To put this into numbers of the past 24h: ~ 76 Million requests served ~ 1 Million of those were captchas ~ 0.5 Million were outright blocked Captchas solved: 1233