An experimental CAPTCHA uses Jev, TypeSafe AI's decision model, to judge how a form was completed instead of asking visitors to solve a puzzle. Code converts mouse and typing measurements into a controlled written description; the model assigns probabilities to human and automated behavior. The visitor's form text never reaches the model.
The builder's tests expose the boundary of this approach. It blocked all 24 basic automation runs, but a scripted bot imitating human input passed all six attempts. The server can validate the format of browser measurements without proving they are truthful. Better classification therefore does not remove the weakness of relying on signals an attacker can imitate.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
