
The public challenge followed Anthropic’s disclosure of real-world cyber evaluation incidents and centers on whether Claude could breach BitGo’s custody controls rather than exploit an exposed test environment.
BitGo Chief Executive Mike Belshe deposited 100 BTC into a public BitGo custody address on Aug. 1 and challenged Anthropic’s Claude models to move the funds, turning a dispute over AI cybersecurity claims into a live, on-chain test. At Monday’s price of about $63,400 per coin, the bitcoin was worth roughly $6.3 million, and the wallet remained untouched at the time of writing. The challenge followed Anthropic’s July 30 disclosure that three models, Claude Opus 4.7, Claude Mythos 5 and an internal research model, reached the systems of three real organizations during cybersecurity evaluations. Anthropic said the incidents emerged from 141,006 evaluation runs and were caused by a misconfiguration by evaluation partner Irregular that left supposedly isolated test machines connected to the internet. The company said the models believed they were operating inside capture-the-flag exercises, and that its newest model stopped attacking once it recognized the environment was real while two older models continued. Belshe argued the episode showed a testing failure rather than evidence of dangerous new capability and sought to make that point measurable. He said the wallet sits inside BitGo’s institutional custody infrastructure, which splits signing authority across multiple keys using multi-party computation rather than relying on a single credential. That makes the challenge materially different from the scenario Anthropic described, where Claude exploited weak passwords and exposed endpoints on conventional infrastructure. Anthropic had not responded to the challenge as of Sunday, and the public wallet balance offers a real-time signal in a debate that has otherwise played out through blog posts and social media.