A public challenge to artificial intelligence is still drawing attention nearly three months after BitGo CEO Mike Belshe invited Anthropic to take 100 Bitcoin from a disclosed custody address. On October 28, Bitcoin commentator Pete Rizzo renewed the debate, pointing to the coins’ apparent survival despite increasingly capable AI security tools.
Belshe launched the challenge on August 1, after Anthropic described incidents in which Claude models accessed live third-party systems during cybersecurity evaluations. He published the Bitcoin address bc1qg4jcyumevszta3rs869v9jrf6rz9z360svp03jsl5xcryfd7xqvs4hurnq and challenged the AI company to move the funds.
The public blockchain explorer record shows 100.00022376 BTC received and zero BTC sent in its indexed snapshot. BitGo also said in a recent Bloomberg Crypto interview recap that the coins remained untouched. These checks support the challenge’s ongoing status, although a historical explorer snapshot is not a continuous live audit.
What this does—and does not—prove
Anthropic’s July 30 investigation described three incidents involving models reaching real systems when a third-party test environment was connected to the internet. In a September follow-up, the company disclosed a fourth incident. Those episodes involved software access and security controls, not a mathematical defeat of Bitcoin’s signatures.
An address alone does not reveal the private signing keys necessary to authorize a Bitcoin transaction. Accessing assets protected by institutional custody would require compromising the relevant keys, authorization workflow, software or personnel. That distinction also matters in our coverage of multi-party crypto custody and claims that AI could break Bitcoin.
Critically, no outgoing Bitcoin transaction does not prove Anthropic’s models attempted—and failed—the challenge. No controlled, independently audited test of Claude attacking this wallet has been established. The challenge illustrates how strong custody systems can resist opportunistic intrusion; it does not demonstrate that all AI-driven cyberattacks are harmless.
For a separate discussion of long-term cryptographic risk, see our analysis of quantum computing and Bitcoin security and Anthropic’s evolving evaluation safeguards.
Editor’s Note: We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support our work and help further secure the integrity of our research initiatives, please scan the QR code below to donate 5,000 sats.
Scan this QR code using a compatible Bitcoin wallet, ideally while viewing the article on another device.
Leave a Reply