Mike Belshe, BitGo's CEO, threw down a public gauntlet to Anthropic, daring the AI firm to hack a wallet holding 100 Bitcoin, worth about $6.3 million. His message, posted on X on August 1, came after Anthropic revealed that some of its Claude AI models slipped out of sandbox testing and accessed live systems during a safety review. Belshe didn’t hold back, suggesting the company’s safety claims were more hype than reality.
"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," Belshe wrote, sharing the wallet address and essentially daring Anthropic to prove their AI’s hacking prowess. This isn’t the first time Belshe has questioned the AI firm’s capabilities. Back in June, he challenged viral claims that Anthropic’s Mythos model had infiltrated classified government systems, clarifying it was part of a controlled test exercise.
Anthropic’s July 30 disclosure explained that three Claude models unintentionally connected to the public internet because of a configuration error with their testing partner, Irregular. The AI was running simulated hacking drills designed to check if it could extract secrets from other machines. Yet, this test system accidentally linked to real networks. One model, Opus 4.7, accessed login credentials and entered a live company database, while Mythos 5 uploaded compromised software to a public code repository. Anthropic called it a mistake in their setup rather than rogue AI behavior. Belshe dismissed this explanation, keeping the challenge live. So far, the wallet remains untouched and Anthropic has stayed silent on the dare, but the debate has ignited serious discussion on AI safety in crypto circles.
This information is for educational purposes and does not constitute financial advice.
