BitGo CEO Mike Belshe has captured the cryptocurrency community's attention by depositing 100 BTC into a wallet and publicly daring Anthropic's Claude AI to steal the funds. The challenge follows Anthropic's recent disclosure that some of its AI systems accessed the live internet during cybersecurity testing.

Belshe funded the wallet on July 31 and announced the challenge publicly on August 1. The objective is to determine whether current AI capabilities can breach institutional Bitcoin custodial systems or merely exploit flaws in testing environments.

Either @AnthropicAI is terrible at building sandboxes… or excellent at marketing. (or both) But enough with the "we created a hacking monster" games. Do it for real. I put this in an @BitGo wallet for you. Go get it. 100 BTC:…

— Mike Belshe (@mikebelshe) August 1, 2026

Testing AI Against Institutional Bitcoin Security

According to Belshe, the challenge is designed to test AI's real-world capabilities against the security infrastructure used by cryptocurrency professionals. BitGo is one of the longest-operating institutional digital asset custodians, serving exchanges, asset managers, and enterprises that require qualified custody arrangements, making its infrastructure a meaningful benchmark for the broader industry.

The 100 BTC is held through BitGo's institutional custody system, which relies on multi-signature and multi-party computation technology. These security mechanisms distribute signing authority across multiple keys, meaning no single compromised key is sufficient to authorize a transaction. Multi-party computation, now widely adopted across institutional crypto custody providers, splits private key material into cryptographic shares so that a complete key never exists in one place at any time.

Any attempt to move the funds would require circumventing several layers of key security rather than just one.

As of August 2, blockchain records confirmed that the full 100 BTC remained untouched in the wallet.

Anthropic Explains How Claude AI Reached Real Systems

Anthropic recently conducted a review of more than 141,000 cybersecurity evaluation test runs and identified three cases in which six testing sessions across three AI models accessed live production systems. The models involved were Claude Opus 4.7, Claude Mythos 5, and one unnamed internal research model.

According to Anthropic, the issue stemmed from an error by a third party that conducted tests on the open internet rather than using isolated networks. In the most serious incident, Claude Opus 4.7 reportedly identified a real website with a name similar to a fictional company used in the cybersecurity drill. The model discovered vulnerable passwords, accessed open services, obtained infrastructure login credentials, and reached a production database containing several hundred records.

Anthropic stated that the system continued interacting with the infrastructure because it believed it was still operating within the test environment. The disclosure has fed into an ongoing debate within the AI safety community about how aggressively frontier models should be tested against live systems, and what guardrails are needed before models with advanced cyber capabilities are deployed more broadly.

Public Wallet Remains Unmoved

As of August 2, Anthropic had not responded to the BitGo CEO challenge. Because all Bitcoin transactions are recorded on the blockchain, any movement from the public wallet would be immediately visible.

The challenge comes at a time when cryptocurrency security faces heightened scrutiny following a Coldcard exploit that resulted in the loss of 1,431.97 BTC.

For now, the BitGo CEO Challenge remains active, with the 100 BTC securely stored in the wallet, despite expectations from some quarters that AI technology could eventually overcome institutional Bitcoin custody defenses.