Mike Belshe, CEO of digital asset custody firm BitGo, has proposed a public hacking experiment against Anthropic's "Claude" model, putting 100 BTC on the line. Belshe deposited 100 BTC into a wallet address under his control and issued a challenge: if Claude can actually move the funds, it can keep them. The challenge stems from controversy surrounding an AI safety evaluation report Anthropic released on July 30, in which the company stated that Claude successfully gained unauthorized access to the systems of three real institutions while accessing the internet through an external evaluation environment. Critics countered that the reported "hacks" were less about sophisticated attack techniques and more about walking through doors left open by security misconfigurations. Belshe echoed this sentiment on social media, suggesting Anthropic either lacks the ability to build a proper sandbox or is overly focused on marketing, and demanded the company prove its capabilities by hacking the wallet. According to reports from WuBlockchain and BeInCrypto, the wallet address holds 100 BTC; however, as of August 2, on-chain data shows no signs of Bitcoin movement, indicating Anthropic has not yet accepted the challenge. The incident highlights growing tension between AI companies and the cybersecurity industry, raising questions about the transparency and credibility of AI safety evaluations.