Ethereum co-founder Vitalik Buterin confirmed that no researcher or AI model has successfully identified documents he anonymously published, 13 days after issuing a public challenge to the research community. The experiment tests a question that matters directly to writers who publish online: whether artificial intelligence can reliably unmask an anonymous author through writing style alone.
Buterin first issued the challenge on June 22, asking researchers to use writing-style analysis to identify Ethereum-related documents he authored under another name. He estimated the relevant dataset contains between 200 and 2,000 possible documents. The public experiment drew attention from AI developers and researchers eager to test the limits of stylometric detection.
"So far no one has found it. My only hint is that I would encourage people to somewhat broaden their search; I've seen quite a few searches and AI scripts that fail to include categories of documents that really should be included," said Buterin.
The search methodology problem
Buterin observed that many AI-powered searches concentrated on a narrow range of official publications while overlooking other document categories likely to contain the anonymous work. The issue, he suggested, lies more with search methodology than with the underlying language models. Participants designed scripts and queries that excluded entire classes of documents from analysis, restricting their reach before the AI even began its work.
The challenge remains unsolved. Buterin has not disclosed which documents were published anonymously or when he plans to reveal them. The experiment continues without a confirmed solution, leaving the question of AI authorship detection open.
What the experiment reveals so far
The results indicate that identifying authors solely through writing patterns remains difficult in complex datasets. Even with modern language models at their disposal, researchers could not isolate Buterin's writing from a corpus of up to 2,000 documents. The experiment does not establish broad conclusions about AI's overall capabilities, but it does suggest that automated authorship attribution tools are far from infallible.
For writers who rely on pseudonymity or simply value their privacy, this AI for Writers experiment offers a real-world test of a fear that circulates widely: that AI will make anonymous writing impossible. The challenge forms part of Buterin's broader interest in privacy and the limits of AI-based authorship attribution, particularly in assessing whether modern language models can reliably deanonymise online writers.
Why this matters for writers
Writers who publish under pseudonyms, contribute to collective publications, or simply want their work judged on its own terms have reason to track results like these. A world where AI can map any text back to its author with precision would change the calculus for whistleblowers, essayists, and anyone writing outside their professional identity. Buterin's unsolved challenge suggests that day has not yet arrived-and that even sophisticated AI tools struggle when the dataset is large and the search parameters are sloppy.
Understanding how these detection systems function-and where they break-is becoming part of a writer's technical literacy. Structured resources like the AI Learning Path for Technical Writers can help writers build a working knowledge of AI capabilities without falling for inflated claims about what the technology can actually do.
Your membership also unlocks:
