Ethereum Co-Founder Vitalik Buterin Unmasked by AI Stylometry in Two Hours

Ethereum News

Ethereum (ETH) co-founder Vitalik Buterin has confirmed a winner in his AI stylometry challenge, after researcher Franklyn Wang traced a hidden anonymous document back to him in roughly two hours. Wang identified a December 2024 revision of EIP-7503, the Zero-Knowledge Wormholes privacy proposal on the Ethereum network, as Buterin’s covert work. Buterin publicly acknowledged the result on Monday with a simple “we have a winner.” The experiment tested whether AI can strip away authorial anonymity, a question with direct stakes for pseudonymous protocol developers who rely on privacy tooling to contribute without exposing their real-world identities.

The contest began on June 22, when Buterin volunteered part of his own anonymity as bait. He offered to reveal one hidden piece of his writing and challenged the public to find it using automated authorship analysis. For 13 straight days, no entrant succeeded. The stalemate suggested that a carefully disguised text could resist machine detection, encouraging developers who assumed that translation and rewriting were enough to stay hidden. That assumption collapsed once Wang applied a purpose-built research engine, turning what looked like an unbreakable disguise into a solvable pattern-matching problem within a single afternoon.

Buterin’s concealment method was deliberate and layered. According to his own account, he first drafted the EIP-7503 revision in Chinese, then translated it into English locally using Alibaba’s Qwen2.5 model, and finally hand-corrected the machine output to remove translation errors. The goal was to scrub away the recognizable rhythm and word choice of his English prose. The rewrite was substantial, ultimately forming roughly 75% of the proposal’s current text. On paper, the pipeline should have severed any stylistic link between the finished document and its author, since none of the surface wording originated in Buterin’s natural English voice.

The decisive weakness was not prose but reasoning. Wang’s analysis flagged that the AI locked onto Buterin’s intellectual habits — his characteristic way of structuring mathematical arguments and explaining algorithms — rather than his sentence style. Buterin conceded the point directly, noting that his obfuscation strategy covered only prose and left his underlying explanatory patterns fully exposed. In other words, translation hid the words but not the mind behind them. That distinction matters technically: stylometry that models argument structure can pierce disguises that defeat vocabulary-based detectors, raising the bar sharply for anyone attempting durable pseudonymity.

The disguise had nearly held under human review. Keyvan Kambakhsh, an original EIP-7503 author, examined the anonymous edit at the time and approved it without recognizing his fellow contributor’s hand. Even the winning AI was not confident: Wang reported that his model assigned the correct pick just 20% probability, yet that figure still ran roughly ten times higher than its next-best candidate. The submission had arrived through a throwaway account, adding another layer of separation. That a low-confidence signal, correctly ranked, was enough to identify the author underscores how thin the margin of anonymity can be against modern tooling.

The result carries implications beyond a single puzzle. Wang argued that the same engine used to identify Buterin, Co-Invest, could be pointed at markets to hunt trading signals across news flow and on-chain data — a claim that positions authorship analysis as one branch of a broader inference toolkit. For protocol developers, the takeaway is starker: reasoning fingerprints may be far harder to erase than writing style, complicating the privacy assumptions baked into pseudonymous open-source contribution. The episode arrives as privacy-focused proposals like EIP-7503 continue to advance, sharpening the debate over how much anonymity contributors can realistically expect.

Turning to the market, COINOTAG’s proprietary 42-indicator composite S/R scoring engine rates the $1,869 resistance at 94/100, its strongest reading, driven by the confluence of the Keltner Upper band and Ichimoku Senkou B, while the $1,747 support scores 79/100 on the Previous Day Low and a bullish Pin Bar. Spot trades near $1,813 as of writing, up 1.75% on the day, with RSI at 58 and MACD flipping bullish despite the broader downtrend. Derivatives lean long: funding sits at 0.0053%, open interest near $6.93 billion, and the long/short account ratio at 1.74 (63.5% long). With the Fear and Greed Index at 24 (Extreme Fear), a reclaim of $1,869 opens $1,982, while a loss of $1,747 invalidates the bullish setup.