BitGo CEO Deposits 100 Bitcoins to Public Address, Invites Anthropic's Claude Model to Attempt Fund Transfer
BitGo CEO Mike Belshe deposited 100 BTC (approximately $6.3 million) to a public Bitcoin address on August 1 and invited Anthropic's Claude model to attempt to transfer the funds.
On-chain data shows that the wallet received funds on July 31, and as of August 2, the balance had not been transferred out; BitGo employs multi-signature or multi-party computation to decentralize signing authority. Anthropic previously identified unexpected interactions between the model and real systems during cybersecurity assessments, but had not responded to this challenge by the time of publication.
In the market, there is growing concern over institutional custody security and the boundaries of AI model capabilities; capital and users are leaning towards multi-signature and other solutions that resist single points of failure, while the risks of purely model-driven on-chain operations are being reassessed, leading to increased demand for custody and security infrastructure.
Source: Public Information
ABAB AI Insight
Mike Belshe has long emphasized that system architecture is superior to implementation details, having publicly discussed multiple times the protection against single points of failure provided by multi-signatures; this large public deposit of BTC and challenge to Claude directly tests the current cutting-edge model's attack capabilities in a real on-chain environment.
On the capital front, BitGo differentiates itself with multi-signature/MPC as a core feature, motivated by the need to demonstrate institutional-grade custody's resilience against AI attacks; this is manifested in publicly verifiable challenges, with funds and trust shifting towards architectures that can withstand model overreach.
Similar cases can be seen in early smart contract vulnerability bounties and public challenges, or security tests of exchange hot wallets; the current crypto security landscape is transitioning from traditional hacker protection to a new phase of countering AI agent capabilities.
This essentially represents a security reconstruction within technological substitution: as AI models gain potential on-chain operational capabilities, market mechanisms are re-pricing custody trust through verifiable challenges, concentrating pricing power towards decentralized keys and robust architectural solutions.
ABAB News · Cognitive Laws
Public challenges are the most direct proof of security
Multi-signature remains central to resisting single points of failure
AI capability boundaries need to be tested with real stakes