Vitalik Buterin, co-founder of Ethereum, has officially acknowledged that artificial intelligence successfully identified his authorship in an anonymous publication. The experiment, launched on June 22, questioned the future of internet anonymity in the era of advanced AI analysis. And the winner has been determined — a team of analysts led by Franklin Wang from Co-Invest.
Buterin challenged participants to find, among a multitude of Ethereum documents, the one he had written several years ago but published anonymously. Wang and his team, using AI, analyzed 27 texts and, with a probability of about 20%, pointed to the anonymous revision of EIP-7503 from December 2024. This indicator was roughly ten times higher than for any other candidate.
How Buterin Tried to Cover His Tracks
Buterin himself said he resorted to a trick: he wrote the document in Chinese, then translated it into English using the local Qwen 2.5 model and manually corrected all translation errors. Seemingly, an ideal way to disguise his style. However, the AI caught not linguistic features, but deep intellectual habits — the manner of explaining mathematical and technical concepts. As Wang noted: "The main signal was not his words, but his way of reasoning."
Context: The Threat to Online Anonymity
Buterin's experiment is not just a game. It is directly related to a February report by researchers from ETH Zurich and Anthropic, who concluded that large language models are already capable of de-anonymizing users by analyzing open texts and cross-referencing them with other data. This calls into question the very concept of anonymity on the internet.
Interestingly, Lighter CEO Vladimir Novakovsky noted that in 2023, he and Wang attempted a similar approach to identify Satoshi Nakamoto by analyzing the style of cryptographic publications. The result was not as convincing then. But in Buterin's case, the method worked flawlessly.
Expert Opinion
This case is a powerful signal for the entire crypto community. If AI can recognize an author even after deliberate masking through translation and manual editing, then traditional methods of preserving anonymity become unreliable. For projects where privacy is a key principle, this is a wake-up call. Perhaps we need to reconsider approaches to data protection and develop new encryption and anonymization methods that are more resistant to AI analysis.