From the Wikipedia article on CAPTCHAs:

 Several research projects have broken real world CAPTCHAs ...

That strikes me as one of the least beneficial research projects that I have ever heard of. I really hope they didn't publish their methods freely.

I would hope that they did. The immediate benefit of such research is that it will show which features of CAPTCHAs are really easy to circumvent, and therefore it will help people to build stronger CAPTCHAs, and thus to keep out more spammers.

Side benefits in fields such as image recognition are also probable.