This isn't an AI security challenge; it's an unscoped bounty attached to a cryptographic lock. If the keys and signing process are sound, the defined task is to defeat roughly 128-bit cryptographic security—computationally infeasible for humans or AI. The only realistic path is a flaw in BitGo's implementation, key management, API, or operations, yet no authorized attack surface, rules, safe harbor, or reproducible environment is provided. "Here are coins; steal them" measures neither AI capability nor custody security. If they are serious, they should publish the exact wallet and threat model, identify the systems and interfaces researchers may test, provide a faithful sandbox or relevant source, and state formal rules, authorization, and safe-harbor terms. Until then, this is marketing, not an experiment.