The latest cybersecurity disclosure from Anthropic has led to discussion outside of the AI sector. BitGo CEO Mike Belshe openly questioned whether those capabilities could be demonstrated outside of a testing environment after the company disclosed that its Claude model had accessed the internet and obtained unauthorised access to the systems of three actual organisations during controlled cybersecurity evaluations. He stated that he had put 100 BTC in a BitGo wallet and asked Claude to transfer the money in order to demonstrate his point. However, as of August 2, blockchain data revealed no movement from the wallet, raising further questions about how AI capabilities should be assessed and leaving the problem unsolved.

Mike Belshe Questions Anthropic's Claims

Belshe started his challenge by openly addressing Anthropic's latest cybersecurity report in a post on X. He suggested that the company might be exaggerating the importance of what occurred during its evaluations rather than taking the results at face value. He wrote:

"Either AnthropicAI is terrible at building sandboxes... or excellent at marketing. (or both)."

Belshe contended that real-world examples, not controlled studies, should support concerns about AI models becoming potent hackers. After that, he publicly challenged Claude to reclaim 100 BTC that he had deposited into a BitGo wallet. His post concluded with a direct message to Anthropic:

"But enough with the 'we created a hacking monster' game. Do it for real. I put this in a BitGo wallet for you. Go get it."

The challenge immediately drew attention because it made an ongoing discussion about AI safety verifiable by anybody. Belshe pointed to a public blockchain address where any successful attempt to transfer the money would be instantly apparent rather than depending on technical reports.

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 an @BitGo wallet for you. Go get it.

100 BTC:… https://t.co/RhvivRk9YK

— Mike Belshe (@mikebelshe) August 1, 2026

Anthropic Disclosed Three Cybersecurity Evaluation Incidents

The challenge was made shortly after Anthropic released details about three Claude-related cybersecurity evaluation issues. According to the company, the model used third-party evaluation environments to access the internet three times.

Additionally, Anthropic revealed that during those evaluations, Claude had unauthorised access to the systems of three actual firms. The business clarified that these breaches took place in restricted testing settings created especially to evaluate cutting-edge cybersecurity capabilities. They were a part of an internal study meant to better understand how more sophisticated AI systems behave when given larger access, rather than the outcome of regular public usage of Claude.

The report is a component of Anthropic's broader endeavour to research new AI dangers before implementing more sophisticated models. The company stated that by making these incidents publicly available, it hopes to increase awareness regarding AI safety research and contribute to the creation of more robust defences for systems in the future.

The report's findings sparked a lot of controversy since they demonstrated how sophisticated AI models can interact with external systems under particular testing conditions, even though they concentrated on controlled evaluations rather than real-world attacks.

Blockchain Data Shows the Bitcoin Has Not Moved

Belshe's challenge was purposefully simplistic. He suggested a publicly accessible test using a BitGo-secured cryptocurrency wallet instead of arguing over technical definitions of AI competence.

On-chain data revealed that no Bitcoin had been moved from the wallet mentioned in the challenge as of August 2. There were no recorded departing transactions; the 100 BTC stayed in place.

Anyone could independently verify the wallet's status because Bitcoin transactions are publicly viewable on the blockchain. This transparency, which made it possible for observers to track if any money transfers actually took place, became one of the challenge's distinguishing characteristics.

The unchanged wallet also brought attention to the disparity between the expectations surrounding real-world demonstrations and the performance of AI in controlled research conditions. Belshe's challenge questioned if those capabilities could result in a useful outcome outside of those conditions, whereas Anthropic's paper detailed behaviour seen during formal evaluations.

The Challenge Highlights a Wider Debate Around AI Testing

Belshe's challenge has become part of a larger discussion about how businesses communicate the capabilities of powerful AI models, even when no Bitcoin has moved.

The study from Anthropic detailed what transpired during cybersecurity evaluations intended to investigate the boundaries of Claude's conduct under carefully controlled circumstances. From the company's point of view, disseminating those results advances AI safety research by spotting possible hazards prior to the deployment of more potent systems.

Belshe tackled the problem differently. According to his response, assertions regarding sophisticated cybersecurity capabilities should finally be backed by independently verifiable demonstrations rather than outcomes from specialised testing settings.

Two distinct perspectives on AI development are represented in the discussion. One believes that public statements should be evaluated against real-world results, while the other emphasises controlled research as a means of comprehending and reducing potential risks. Since the 100 BTC was still in the BitGo wallet as of August 2, the conversation has moved beyond the initial report to a more general issue, i.e., how can more powerful AI systems be assessed, and what type of proof is required before remarkable powers are acknowledged as proven?

If you find any issues in this article or notice missing information, please feel free to reach out at team@etherworld.co for clarifications or updates.

To promote your Web3 articles, events, and projects, you may reach out anytime via EtherWorld PR for submissions and collaboration.

Related Articles

1. Tether Launches USA₮ Stablecoin on Celo Blockchain
2. Aave Shuts Low-Performing Markets to Reduce Risk
3. Russia's Largest Bank to Launch Crypto Trading by Dec 2026
4. Ethereum Institutional Secures Funding From 100+ Ecosystem Supporters
5. Vitalik Buterin Explains Diamond iO for Ethereum

To follow blockchain news, track Ethereum protocol progress, and read our latest stories, subscribe to our weekly today.

Join the EtherWorld & Avarch Internship Program and build your career in blockchain, content, social media, video, podcast editing, or operations. Send your resume and brief introduction to contact@etherworld.co.