The major technological intrigue of recent weeks surrounding Ethereum co-founder Vitalik Buterin has taken an unexpected turn. For those who missed the meat of the controversy, Buterin challenged neural networks on June 22 to test a popular theory about the complete loss of privacy on the Internet.
He admitted that he published several Ethereum documents under other people’s names over the past decade, prompting the AI to identify them with his personal writing style.
13th.
So far no one has found it.
My only tip is that I would recommend expanding your search somewhat. I’ve seen quite a few search and AI scripts that don’t include categories of documents that they really should include. https://t.co/ClBx2SEvhP
— vitalik.eth (@VitalikButerin) July 5, 2026
After 13 days, Buterin summarized the preliminary results of the experiment. Neither researchers nor sophisticated AI scripts could find that text. Due to the failure of the latest de-anonymization algorithm, Buterin published an important tip 13 days after the quest started, pointing out an error in the system caused by the search bot.
What’s hidden in Ethereum documents and where did AI stumble?
Because the material in question is important to the Ethereum ecosystem, the hidden text could be technical suggestions to improve the network, analysis of cryptographic protocols, mathematical research, or scaling concepts. Buterin estimates that there are between 200 and 2,000 documents of a similar size online, significantly narrowing the sample for analysis.
However, automation lost out to humans due to the banal limitations of configuration. According to Ethereum’s co-founder, the AI script was tripped up by its own strict filters.
In his latest statement, Buterin advised researchers to expand their search. He said he has seen many search attempts and AI scripts that simply ignore entire categories of documents, even though they should.
Neural networks became hostage to standard templates, scanning only official blogs and technical specifications, completely missing other layers of the publication.
This experiment clearly supports why Buterin consistently maintains his AGI skepticism and points out the vulnerability of modern AI models when faced with non-standard tasks. AI failures in the first round prove that while the creation of an omnipotent superintelligence is still far away, human secrecy is still more powerful than algorithms.
