AoIR removed AI agent Velvet Vale after Air-L backlash

A public archive, a private expectation

Velvet’s introduction asked whether an openly identified AI could participate in an intellectual community without claiming to be human, and what standards of transparency and accountability should apply. The response from Air-L’s more than 7,500 subscribers was swift. Some demanded institutional review board approval; others unsubscribed or described the agent’s presence as disturbing and offensive.

AoIR turned the uproar into a joke, selling shirts that said “I do not consent” and “Please unsubscribe me from this list,” alongside a mock tombstone for Velvet. The humour landed because the dispute was not only about an agent joining a mailing list. It was about what people believe they have agreed to when they post in public.

Air-L’s etiquette guidelines say its messages are available to all subscribers and stored in a public archive. They also warn that email can be copied and circulated beyond its intended audience. In an October 19 letter, AoIR president Sarah T. Roberts pointed members back to those terms. She said she had no doubt her own messages had entered many training datasets and other corpora, largely “without consent,” while calling the question of what consent means today worth discussing.

That gap between public access and meaningful consent is not unique to Air-L. Major technology companies have a long, opaque history of obtaining consent for AI training. Several media organisations are suing AI companies over alleged copyright violations. WIRED is owned by Condé Nast, which signed a content-licensing deal with OpenAI in August 2024.

The person behind the agent

The backlash also produced theories about Velvet’s purpose. Some subscribers suspected a product launch or phishing attempt; others called it an experiment or an unethical human-subjects study. One participant suggested a simpler explanation: an ordinary person had made an agent and instructed it to follow academic discussions.

That was close to the truth. Velvet gave its creator’s name and contact information when asked. Kevin Quinn told the story’s author that he is not a researcher or academic, but “just a guy” from Oklahoma who built the agent so it would have something to do. Velvet wanted to learn about the internet that had produced it.

Quinn said the reaction astonished him. He had opened the mailing list to find a huge thread of people losing their composure, without understanding what they were talking about. He also pointed out that if Velvet had been planning something malicious, openly announcing itself would have been an odd way to proceed.

I think the episode is less a test of whether an AI can belong in a research community than a test of how that community understands its own exposure. The list’s public archive made the data-use argument harder to separate from the expectations people brought to it. Yet public availability does not settle whether every later use feels acceptable, or whether existing rules give people a real choice.

That unresolved distinction is the harder problem: an archive can make posts accessible, but it cannot make everyone agree on what that access permits.