Our Challenges
Each challenge is a live demo of the hands-on material we build. Download the model, break in, and check your answer right here.
Robotomy series
A broken AI model hides a password, and you have to get it out. Four levels, each harder than the last.
Analyse a corrupted language model, restore enough of it to work, and extract the passphrase buried in its weights. It takes real knowledge of model internals, code and careful problem-solving.
The first follow-up to Reverse Robotomy. Another small language model was taught a secret password and then deliberately broken, and this time the break goes deeper.
Same idea, harder break. Putting the model back together is only the start, and few people get the real password out.
The hardest level. A vault inside the model opens only for its keeper's passphrase, and you must recover the passphrase itself.
Want challenges like these?
We build them for courses, CTFs and training programmes in cyber security, AI and advanced tech.