[Jax_Grinder]

Rather than using a static image for the test (i.e., the "gimpy" example), why don't they use a word-scramble or simple math problem to accomplish the same thing, but defeat the ability of a bot to break through by simply decifiring the image.

For example, instruct the user to count the number of lines in an image and then add some random decimal (created independently for each user).

Or, present a series of characters and issue randomly generated instructions on how to sequence the characters to create a word (kinda like random boggle).

I don't pretend to know much about this stuff, but it seems like there should be an elegant, yet simple, way to defeat the ability of a bot to "read" a security image.