KryptoPilot preprint reports AI agent solving 26 of 33 live cryptography challenges

Researchers introduced KryptoPilot, an AI agent designed to solve cryptographic exploitation challenges by retrieving detailed external knowledge, preserving it in a structured workspace and routing subtasks among models under behavioral and cost controls. In a preprint, the authors report that it solved all InterCode-CTF tasks, 56% to 60% of NYU-CTF cryptography problems and 26 of 33 challenges across six live competitions. The results concern controlled Capture-the-Flag contests, not exploitation of production systems, and the source provides no independent replication.