A revised EIP-7503 draft written in Chinese and translated with Qwen2.5 was still linked back to the Ethereum co-founder, highlighting how AI may detect authors through technical reasoning patterns.
Vitalik Buterin said researcher Franklyn Wang won his AI authorship challenge after tracing an anonymous December 2024 revision of EIP-7503, the Zero-Knowledge Wormholes privacy proposal, back to the Ethereum co-founder. Buterin launched the experiment on June 22 to test whether AI stylometry could identify his hidden writing, and no one succeeded for 13 days. Wang used Co-Invest, an AI research engine, to flag the rewrite, which was submitted through a throwaway account and now makes up roughly 75% of the proposal’s text. The anonymous edit had initially escaped detection even though Keyvan Kambakhsh, an original EIP-7503 author, reviewed and approved it at the time. Wang said his model assigned the match 20% confidence, though that was 10 times higher than its next candidate, and the search reportedly took about two hours. Buterin said he had drafted the revision in Chinese, translated it locally with Qwen2.5, and then manually corrected the output. He wrote that AI still identified him through his math and algorithm explanation patterns, suggesting the system picked up deeper reasoning habits rather than surface prose. The episode highlights a growing tension for a crypto industry long shaped by pseudonymous builders and privacy concerns, and may add to debate over how far increasingly capable AI systems can go in deanonymizing anonymous work.