A moderation bot posted 200 times by mistake and helped give Usenet “spam” its name
Retro-moderation already had critics. Its automated cleanup then delivered a very public demonstration of what could go wrong.
Richard Depew’s ARMM software was supposed to help moderate Usenet after messages appeared.
On March 31, 1993, it posted 200 messages in a row to news.admin.policy, the group where people discussed running the network. In Templeton’s account, Joel Furr was the first person he could establish had called that flood spam. Templeton qualifies the priority with “as far as I can tell”. Depew subsequently used the term in an apology.
The tool meant to remove disruptive material had become disruptive material. The surviving uncertainty concerns the earliest use of the word, not an absence of earlier unwanted messages.
Depew’s proposed moderation happened after publication. Ordinary moderated groups screened a message before readers saw it; his retro-moderation idea would allow messages through and cancel rule-breaking ones afterward. Templeton describes the proposal as controversial because canceling someone else’s posts was normally a serious breach of etiquette. The software’s job already touched a sensitive boundary before it began flooding the discussion.
The destination made the accident particularly visible. news.admin.policy was where people argued about operating the network, so the bot brought its failure directly to the people debating that authority. Two hundred repeated posts gave them a fairly immediate example of what could go wrong with an automated participant.
Templeton separates several earlier events that are easy to collapse into a single origin myth. A 1978 DEC email advertisement predates this episode; a mass Usenet plea for money appeared in 1988. The unwanted behavior existed before the term became familiar in that setting. Dating a word and dating an annoyance are separate research problems.
His account follows the word through MUD communities as well, where repeated chat text or automated floods could be called spam. That history makes the ARMM episode a meeting of an existing expression and a memorable failure. It does not require somebody to have invented the word while watching the bot malfunction.
For a present-day operator, the recognizable risk is the permission to create more output in response to events. A duplicate event should not become permission for another unlimited sequence of posts. Testing that behavior in a private destination, with a small output allowance, is a direct way to see whether the stop mechanism works before the audience becomes the alarm.
A recovered message could revise who used the word first. It would take different evidence to change what the bot did that day, which is why the qualification belongs beside the priority claim.
Automation that can publish needs a way to stop publishing. The next newly recovered archive could still move the word’s earliest documented appearance further back.