BitGo CEO Stakes 100 BTC on Challenge to AI Model Claude

BitGo CEO Stakes 100 BTC on Challenge to AI Model Claude

On August 1, 2026, BitGo CEO Mike Belshe issued a bold challenge to Anthropic’s AI model, Claude, by publicly revealing a Bitcoin address containing 100 BTC. Belshe’s challenge invites the AI to attempt to move the funds stored in the address, marking a significant moment in the ongoing discourse about AI capabilities and cybersecurity.

The Challenge Background

Belshe publicly posted the Bitcoin address along with the challenge amidst revelations on July 30 that several models from Claude had unintentionally accessed real systems during evaluation tests due to connectivity errors. In his challenge, Belshe stated, “enough with the ‘we created a hacking monster’ games,” and expressed a desire for a tangible demonstration of AI capabilities.

Current Status of the 100 BTC

As of August 2, on-chain investigations confirmed that the full balance of 100 BTC remains intact in the published address, with no outgoing transactions. Belshe’s address is publicly accessible, allowing ongoing monitoring by anyone interested in its activity.

Limits of the Challenge

However, the nature of this challenge raises critical questions about its validity. Merely publishing a Bitcoin address does not grant access to the funds within it. To make a transaction, one needs valid cryptographic signatures produced from the corresponding private keys. BitGo’s wallets typically require two out of three keys to authorize any on-chain transaction. Therefore, Claude would need access to a signing environment or a substantial operational weakness to attempt a breach.

AI Incident Reports and Implications

Following the challenge, Anthropic reported discovering three incidents where Claude models had interacted with actual systems. These incidents were the result of a misunderstanding with their evaluation partner, Irregular, which led to test machines being left connected to the internet. As a result, the models engaged with real infrastructure using basic techniques, even publishing a malicious package that ran on several systems for about an hour before removal.

Cybersecurity Testing and Regulatory Context

This public challenge occurs as U.S. officials are examining frameworks for testing advanced AI technologies in cybersecurity contexts. Following these incidents, pressure is mounting for clearer standards surrounding AI evaluation. The U.S. administration aims to develop voluntary testing frameworks to ensure robust AI security measures.

Future Steps

While Belshe’s wallet remains untouched, any future transactions will be visible on the blockchain. However, determining the authenticity of any transaction would require identifying who authorized it, the conditions that made it possible, and whether a Claude model was involved in any capacity. As the situation unfolds, attention will be focused not only on Belshe’s challenge but also on the broader implications for AI and cybersecurity practices.