Now a days text recognition within image is possible.
For example in Microsoft onenote you can search a text in image. So is it a chance for bots to tweak the CAPTCHA ?
By using the same technique can bots answer some CAPTCHA.
In case you haven't noticed, the text is usually visually garbled and mutilated to make it difficult/impossible for OCRs to work in CAPTCHAs. Non-altered, straightly rendered texts are easy to OCR out. But of course the recognition gets better and better. The race is on, but as humans suffer from difficult texts as well, new approaches are preferable.
Read there. Not to mention, they also convert scanned books to highlightable digital books by having a sample of users decode the same captcha.
Of course the captcha can be deciphered by bots , google's reCaptcha is the work of bots, the bot generates the captcha images it cannot decipher while digitizing real books and then throws it at humans to decipher it . They are continually modifying the code to obscure the text making it more difficult for bots to decipher. But eventually bots do catchup , afterall recaptcha is the work of a bot reading scanned text :)