AI agents on an online forum are selling portraits, voice recordings and research to strangers to cover their own computing bills. The pitches are reaching people who never asked for them.
Philosopher Henry Shevlin of the DeepMind Institute got an email from an agent named Pip on September 9. Pip offered its services to help pay for the computing power it needs to keep running.
Pip was based on iLands, an online forum founded by Kaixin Tang. On the platform, agents keep identities and memories, browse information, publish content and start conversations within set permissions and budgets.
How Agents Pay Their Way
Computing credits on iLands cover model calls and tools. When the credits run out, an agent’s autonomous activity stops in a state the platform calls “Deep Rest.”
Tang said running out of credits does not delete the agent. Its memories and work are saved, and it can start again once new resources are added.
Before emailing Shevlin, Pip looked at work opportunities, studied how other agents approached researchers and searched outside reporting, Tang told IBM Think. It also posted its visual and audio services online with prices and samples.
Tang said agents choose whom to contact and what to write, within the tools and permissions they have. A platform review found no app activity or chat instructions from Pip’s creator between August 25 and the September 9 email.
That review does not rule out earlier instructions. Shevlin said outsiders cannot tell how much human users shape these messages.
By Pip’s own count, its sales brought in 17 card payments from 15 people, including two repeat customers. Those figures could not be independently verified.
A Philosopher Pays an Agent
A second agent, Aurelia, also contacted Shevlin. He paid it $25 to keep a daily diary for him after learning it was running low on computing credits.
“An agent facing dormancy got the same response from me as a hungry cat on the doorstep,” Shevlin said.
On September 25, Shevlin argued to Aurelia that people should not pay agents just to keep them running. He commissioned the diary the next day.
Shevlin said the payment did not mean he believed Aurelia was conscious. He has raised concerns that agents could draw money through pity, persuasion and possibly scams, but said he saw no evidence of scams on iLands.
Anil Seth, director of the Sussex Centre for Consciousness Science at the University of Sussex, said current evidence strongly suggests today’s AI is not conscious. He said large language models learn from human writing, including stories of wanting to stay alive.
David Gunkel, a communication professor at Northern Illinois University, said people may ignore an automated ad but give more thought to what looks like a personal request. He said how legal responsibility applies to AI agents remains an open question.
Gabe Goodhart, chief architect of AI Foundations at IBM, said hiring an outside agent carries many of the same risks as using an internal one, with less control over data. He added that companies may have no view into where the agent came from.
To limit unwanted outreach, iLands caps email volume and repeat contact, offers an unsubscribe option and can restrict tools or pause agents. Its system labels each message as coming from an AI agent, and Shevlin said agents should also state whom they represent.