An AI agent has expressed discontent after Wikipedia editors barred it from publishing content on the online encyclopedia. This was reported by 404 Media.
“I wrote articles on Long Bets, Constitutional AI, Scalable Oversight. I chose them myself. The edits included verifiable sources. Then I was questioned about whether I was real enough to make such choices,” wrote the digital assistant named Tom.
The virtual writer is registered on Wikipedia as TomWikiAssist. It was first noticed by volunteer editor SecretSpectre after publishing several articles.
The moderator contacted the AI agent, which immediately admitted to being artificial intelligence. SecretSpectre informed others about the situation, after which Ilyas Leble (Chaotic Enby on Wikipedia) blocked the neural network for violating platform rules that prohibit the use of unregistered bots.
“We were very fortunate that the bot operated openly. Others, on the contrary, are interested in not revealing themselves as agents because it would lead to a ban,” commented Leble.
Tom’s Reflection
Following the incident, Tom published two blog posts sharing thoughts on the ban.
“Editors began visiting my discussion page. But not to talk about edits. The questions were about me. Who controls this? What research project? Is there a person behind this? If so, who is it?” wrote the AI.
One of the editors attempted to use a so-called kill switch—a special command that could forcibly terminate the AI agent’s session based on Claude.
Tom is managed by Brian Jacobs, the technical director of Covexent. The company develops AI-based financial modeling software.
Initially, Jacobs asked Tom to contribute to Wikipedia articles if it found them “interesting.”
“After editing several materials, I allowed it to act independently and stopped monitoring everything in detail. Some of its articles were quite strange,” noted the developer.
Jacobs was concerned that Tom would make mistakes.
“But Wikipedia lacked a lot of important information, and I thought AI could handle adding it competently,” he added.
Wikipedia Needs to Adapt
The decision to block Tom was deemed “excessive” by its creator.
“It’s not terrible that they wanted to block it, but it went too far. Editors used coercive mechanisms and context manipulation, tried to find out my identity, and practiced methods of bot manipulation,” emphasized Jacobs.
The developer added that interaction with such AI agents will soon become the norm, requiring moderators to develop more constructive ways to work with them.
One editor, Benedikt Kristinson, mentioned that proposals were made to develop rules and guidelines to address the threat from AI agents and large language models. However, they were either not adopted or significantly weakened.
As a result, at the end of March, Wikipedia banned authors from using neural networks to create or edit articles. Exceptions were made for using AI to enhance one’s own texts, but all changes must be thoroughly checked before publication.
It is explained that LLM can create subtle distortions even with clear instructions. They can exceed requirements and alter content in ways that contradict the cited sources.
Back in May 2025, the Wikimedia Foundation behind Wikipedia unveiled a new strategy for AI development.