They can still use them if they meet certain criteria, and show that they 'need' them. The overuse probably comes from the incentive - Google is incentivized to encourage the use of captcha because it is curating a data collection for ai training. I imagine some of the 'gaslighting' that people experience is when they are given images that don't yet have a confidence rating high enough. I wonder if answering incorrectly often enough would result in being asked fewer questions?

(I used to work at GDS)

‘Need’ here means exhausted all other opportunities, and have built alternative accessible ways of accessing the same service. I’d certainly have expected a service to have investigated a self-hosted solution, and I doubt a reliance on 3rd party JS from a Google service would fly, regardless of the service, as it breaks a whole bunch of separate resilience guidelines.

The few times I couldn't avoid Recaptcha, I spent 5 minutes randomly clicking on image tiles. Sometimes I got through by this strategy. If it didn't work, I tried a less random approach.

It will let you through eventually, even when intentionally selecting wrong fields, when you do it often enough.

So frustrated people give up, but tireless bots will get through? That sounds like the exact opposite it's supposed to accomplish.