Captcha?

Based on " Drawing lessons in notepad ". From this you can make a captcha. And almost impenetrable. The image that needs to be recognized is in the body of the html document, which means you have to either visualize it completely, which can be a lot (if the forum, for example, there are higher / lower messages), or select the DOM subtree and visualize separately. A simple trick with css will cause the selected subtree to not look right. In addition, the code can be “smeared” throughout the document, and then the DOM subtree will not do the trick (by the way, the selection of the necessary DOM element by the set of conditions is already an AI problem, not always successfully solved).

If the java-script is connected to the case, the robot will also need the ability to execute it.

And only after going through all these steps will come the turn of the OCR. There is a thought that not every robot will get to this stage ...
')
UPD.

Drew an example . By the way, this captcha slightly exceeds the usual security level. Distorted text (the degree of distortion is small on purpose) is easily cut out in at least 2 ways.

Of interest may be a combination of drawn text with a question for understanding. It is impossible to answer the question without reading (or listening to) a fragment of text in which a picture created in this way is embedded.

There are several tasks for the robot: to find the recognition area, to recognize the necessary fragment of text and to generate the answer to the question. The main thing is to make it impossible to do without recognition. Those. get a high enough guarantee from creating an answer selection algorithm.