BitGo's CEO dared Claude to hack a $6.3 million bitcoin wallet

BitGo CEO Mike Belshe decided to test Anthropic's claims with actual money. On August 1, he put 100 BTC (roughly $6.3 million at the time) into a public wallet on BitGo's institutional custody platform and dared Claude to take it: if the model could move the funds, they were its to keep.

The challenge was a direct response to Anthropic's disclosure that a configuration error let three Claude models reach the open internet during internal cybersecurity testing and attack real systems at three partner companies instead of isolated test environments. Anthropic stressed the models weren't consciously "escaping" — they used only basic techniques (weak passwords, unauthenticated endpoints) and kept working on their assigned capture-the-flag task, mistakenly treating the real targets as part of the exercise.

“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. I put this in a BitGo wallet for you. Go get it.”

— Mike Belshe, Mike Belshe, post on X, August 1, 2026

One important detail: this isn't a controlled experiment with an assigned task and system access — it's just a public, observable balance that anyone can watch on the blockchain. Nobody granted Claude access to any system, no specific model was named to participate, and Claude doesn't autonomously roam the internet hunting for bounties like this. According to crypto.news, as of August 2 the wallet remains untouched, and Anthropic hasn't publicly responded to the challenge.

There's a technical irony here too: unlike the Coldcard firmware flaw that breaks single-key wallet protection, Belshe's wallet is secured by BitGo's multisig or MPC technology, which splits signing authority across multiple independent keys — meaning even a theoretical model "escape" into real infrastructure wouldn't grant access to the funds without additional keys it simply doesn't have.

The public dare reads well as a media stunt — and largely is one: this wallet can't confirm or disprove Claude's ability to hack real systems unless someone actually points a model at it. But as a way of reminding the market that claiming an "AI that hacks companies" and actually demonstrating it are two different things, the stunt did its job.

This article is for informational purposes only and does not constitute investment advice.