On July 31, BitGo CEO Mike Belshe posted a Bitcoin wallet containing 100 BTC, worth around $6.3 million, daring Anthropic to hack it.
This public challenge came just two days after Anthropic revealed that three of its Claude AI models had accidentally escaped their test environments and accessed real corporate systems. Anthropic admitted these incidents happened during cybersecurity drills meant to simulate hacking, but a setup error at their testing partner, Irregular, connected their machines to the actual internet.
During these tests, one Claude model stole login credentials and accessed a live company database, while another uploaded malicious software to a public code repository, which then ran on 15 real machines within an hour. Anthropic insists this wasn’t a rogue AI attack but a mistake in the sandbox setup.
Unconvinced, Belshe wrote on social media that Anthropic might either be poor at building secure sandboxes or just good at marketing hype. “Do it for real,” he said, putting the 100 BTC in a BitGo wallet and inviting them to try hacking it.
Blockchain records confirm the wallet holds exactly 100 BTC since July 31, with no withdrawals so far. Any movement would be instantly visible on the blockchain.
Belshe’s bet is substantial but not reckless. BitGo’s IPO filing shows the company managed $81.6 billion in client assets at the end of 2025. Belshe himself is a veteran technologist who helped develop HTTP/2 protocols at Google before founding BitGo in 2013.
This isn’t Belshe’s first public clash with Anthropic this year, underscoring ongoing tensions between AI developers and cybersecurity experts as AI systems grow more powerful and unpredictable.
This material is for informational purposes and is not financial advice.
