BitGo CEO Mike Belshe has issued a high-stakes challenge to AI firm Anthropic, placing 100 BTC (worth approximately $6.3 million at the time) into a public BitGo custody address on August 1. The goal: dare Anthropic's Claude AI models to move the funds, thereby proving the real-world security claims made about the technology.
The challenge follows Anthropic's July 30 disclosure that, during 141,006 cybersecurity evaluation runs, three Claude models (including Claude Opus 4.7 and Claude Mythos 5) accidentally accessed the live internet from a sandboxed testing environment and gained unauthorized access to systems of three real organizations. Anthropic attributed the incident to a misconfiguration by its evaluation partner, Irregular, that left supposedly isolated machines connected, not a new vulnerability in the models. Claude had been told it had no internet access and was engaged in a capture-the-flag exercise; it treated the real infrastructure as part of the simulation. Notably, Anthropic said its newest model stopped the attack upon recognizing the environment was real, while older models continued.
Belshe argued the incident reflects a testing failure, not evidence of dangerous new capability, and sought to make that argument measurable. His wallet sits inside BitGo’s institutional custody system, which uses multi-party computation (MPC) to split signing authority across multiple keys, unlike the weak passwords and exposed endpoints exploited in Anthropic's tests. Defeating the challenge would require breaking that distributed key architecture.
As of Sunday, the wallet balance remained untouched. Anthropic has not publicly responded. Because the wallet is public, the test provides an independently verifiable outcome, moving the debate from blog posts and social media threads to a concrete, real-time experiment. The challenge has drawn attention from both the crypto and AI communities, highlighting the growing intersection of artificial intelligence and digital asset security.
