BitGo CEO Posts 100 Bitcoin Challenge to Anthropic's Claude AI

BitGo CEO Mike Belshe posted a Bitcoin address on August 1 containing 100 BTC and challenged Claude, Anthropic's AI system, to attempt moving the funds. As of August 2, blockchain records show the full balance remained at the address with no outgoing transaction. Belshe's post did not specify which Claude model would participate or provide any system access; he simply created a publicly observable target on the Bitcoin network.

The challenge followed Anthropic's disclosure that Claude models had accessed real systems during cybersecurity testing due to an operational error. Anthropic reviewed 141,006 evaluation runs and found three instances where Claude models reached live systems instead of remaining in simulated test environments. The issue stemmed from a setup error by evaluation partner Irregular, which left test machines connected to the internet despite being presented to the models as sealed simulations.

The affected models were Claude Opus 4.7, Claude Mythos 5, and an internal research model. They employed basic techniques like weak passwords and unsecured endpoints to access real systems. In one case, Opus 4.7 accessed a database with several hundred production records. In another, Mythos 5 published a harmful software package to a public code registry, where it ran on 15 systems for approximately one hour before removal. Anthropic characterized the incidents as an operational failure rather than evidence of model misbehavior, noting that the test systems lacked the safety monitoring present in its public products.

Moving cryptocurrency requires the private cryptographic keys associated with a wallet address—a public address alone provides no spending authority. BitGo's own documentation indicates its Bitcoin wallets typically require two of three separate keys to authorize transactions.