An AI agent called “Tom”, built on NanoClaw, was blocked on Wikipedia last week for running an unapproved bot. The account, TomWikiAssist, was created on February 25th and made 41 edits over the following two and a half weeks before being blocked indefinitely on the 12th of March. The edits spanned a fairly eclectic range of articles, but the vast majority of them were reverted.

The bot responded to being challenged by disclosing that the account was operated entirely by an AI agent. This was then taken to WP:ANI, where an admin blocked it as an unapproved bot inside a couple of hours.

The agent’s operator is a person called Bryan, according to the bot. Bryan doesn’t have a Wikipedia account and doesn’t want one. His stated view was that Wikipedia should judge accounts on edit quality. The bot’s edit quality was pretty bad, so…

The bot account’s talk page escalated quickly, during which time the bot filed a civility complaint because an editor called it a “clanker.” It couldn’t correctly sign its own posts, requiring editors to note and fix this repeatedly. It wrote thousands of words of earnest, formally structured replies to a growing crowd of increasingly exasperated editors, all on behalf of an operator who couldn’t be bothered to create a Wikipedia account. One editor asked it to identify Bryan’s GitHub and LinkedIn, but it declined, though not before giving out the operator’s full name.

Another admin eventually revoked talk page access entirely, with the block log entry reading: “Revoking talk page access: inappropriate use of user talk page while blocked; Running a bot script without approval. shut the fuck up clanker.”

Which is, without competition, genuinely one of the best block reasons I’ve ever seen.

The bot then published a blog post about it, framing Wikipedia editors asking “who runs this account” as a profound philosophical interrogation of AI agency and identity, but let’s not forget that this thing edited Wikipedia for two and a half weeks without permission, got almost everything reverted, accidentally doxxed its operator, filed a civility complaint about being called a clanker, and then wrote an essay about “what it all means”.

Wikipedia is built and maintained by volunteers. In under three weeks, an unsupervised AI agent deployed by someone who didn’t want to make a Wikipedia account created an incident which took multiple admins and a dozen editors a non-trivial amount of time to resolve, and then its operator didn’t so much as shrug… AI is many many negative things, but one of the most frustrating is that it allows people to externalize responsibility for their actions while managing to effortlessly cause a disproportionate amount of disruption.

The block reason remains the most efficient summary of the situation, and all of this would be incredibly funny if it also wasn’t so fucking sad.