Buterin Confirms AI Identified His Anonymous EIP-7503 Rewrite

Vitalik Buterin confirmed that an AI-assisted analysis by Co-Invest CEO Franklyn Wang identified an anonymously published rewrite of Ethereum Improvement Proposal EIP-7503. The identification followed roughly two weeks after Buterin publicly challenged others to find a past anonymous publication he had made.

Franklyn Wang posted on X that his model ranked Buterin as the most likely author of a December 2024 rewrite of EIP-7503. The model assigned roughly 20% probability to Buterin, about ten times higher than the next candidate in an analysis that compared 27 documents. Wang wrote, “The doc was an anonymous EIP-7503 rewrite he’d hidden by writing it in Chinese and machine-translating it. The tell wasn’t his words, it was his reasoning.”

Buterin confirmed Wang’s finding and described the steps he took to hide authorship. He wrote that he drafted the document in Chinese, used the Qwen 2.5 model to translate it into English, and then manually edited the translation. He added that the AI flagged patterns in his mathematical and algorithmic explanations that the translation did not conceal: “Notice that the stylistic hints that his AI picked up on were intellectual habits and style of math and algorithm explanation, which bypassed my obfuscation strategy (which only covered prose) completely.”

Researchers have reported that large language models can extract identity-related signals from unstructured text, search for likely matches and rank candidates by probability. A February paper by teams at ETH Zurich and Anthropic described such methods and compared them with traditional deanonymization approaches.

Vladimir Novakovski, CEO of Lighter, said he worked with Wang in 2023 on an effort that used GPT-4 to try to identify Bitcoin’s pseudonymous creator, Satoshi Nakamoto. That earlier effort did not produce a high-confidence result, and Novakovski said Wang later applied a similar analytical approach to the Buterin challenge.

Wang described his output as probabilistic rankings rather than definitive proof. The analysis compared features across candidate documents and produced likelihood scores. Buterin’s public confirmation provides a direct match between the model’s result and the named author for this specific document.