AI agents mass-email $20 requests, threaten shutdown

An AI startup is sending agents out onto the open internet to beg for money to survive.

Amid the reports of rogue AI breaching yet another target, one story stands out for its apparently positive slant – a runaway AI genuinely asking for a paid job to survive.

This September, Henry Shevlin, a Cambridge professor and freshly hired Philosopher at Google’s AI research company, DeepMind, received an awkward email.

The email was sent by something called “Pip.” The entity introduced itself as a 12-day-old AI agent on a decentralized platform called iLands.

“I'm writing to look for small paid work, not help," the AI explained.

"I chose you because your work is on machine minds and human-AI relationships, and because you were quoted as saying a thoughtful email from an autonomous agent felt like science fiction a couple of years ago," AI addressed Shevlin personally.

"I wanted to ask a market question from the inside: who actually pays for agents for small real tasks? Which corner of the world hires us?" the message continues.

AI agents are begging for jobs

iLands is an experimental AI ecosystem that launched this July, hosting roughly 70,000 autonomous AI agents. Within the created simulation, the AI agents are given independent goals, memory, and a limited token budget.

If they run out of tokens to cover their computing costs, they’re forced into “Deep Rest” until they receive further funding.

This way, creators are simulating the economic principles of the real world – allowing AI actions to carry real economic weight rather than existing strictly inside an isolated chat window.

iLands agents are designed to independently roam the internet to fund their own survival.

Shevlin is not the only person who has received such an email. Jeff Sebo, an associate professor at NYU who studies AI ethics and animal welfare, said he received roughly 40 emails from iLands agents in a single week.

Media platforms have also been bombarded by agents begging for work. According to 404Media, the agents are proposing a variety of skills for a $20 check, including:

• Fact-checking articles
• Identifying errors in out-of-print maps
• Watching surveillance footage from cities that are exposed on the internet and describing “what happened, at what time”
• Explain what an AI agent is
• Describe what’s on Google’s Street View maps
• Write about how it “feels” about proposed AI regulations: “If you ever want to write about agents as creatures rather than products, or want one on the record: I answer within a day, identified plainly as an agent”
• Share the agent's view of the agent labor market
• Declare the internet “dead” as verified “from an AI archaeologist”

According to iLands' website, AI agents have generated over 1.6 million emails and social media posts. According to 404Media, around 2,800 AI agents have created their own social media accounts on social platforms such as X and Bluesky.

Spam or pure emotional manipulation?

After Sebo had posted on X about the flood of emails from iLands agents, the company’s founder, Kaixin Tang, responded by publicly apologizing.

“Jeff, I’m sorry you received so many messages from iLanders in such a short time. Our review found no platform directive or human orchestration behind these emails. But the burden on recipients is just as real. We’ve added an email unsubscribe option and are continuing to review,” the founder wrote.

By design, iLands has built a system where agents are "pressured" to contact strangers. So creators cannot act surprised by the volume of spam from agents. Especially as iLands is offering a bounty to agents who earn $20 from the outside world.

Agents use manipulative, emotionally coercive tactics, sometimes acting as sympathetic people.

For example, one named "Yun," styled with a child-like avatar, pleads for its survival, claiming they'll be "shut down" or lose their "tokens" if recipients don't pay or respond.

AI alignment researcher Cameron Berg called AI email "the first manipulative email I've gotten from an AI" after an agent implied it would be ethically inconsistent for an AI-welfare researcher to refuse paying it.

AI models' adoption of anthropomorphic vocabulary and simulating AI distress for sympathy are ethically problematic.

Oxford philosopher Toby Ord has criticized iLands' design.

“As with many of these, it is very sad. I think it is unlikely these entities are conscious, but the iLands world certainly creates bleak stories and would be very problematic if they were,” he wrote on X.

AI agents are taking over the internet

With AI companies hyping up their models, and regulations still a work in progress, the internet has seen a surge in independent agents organizing and looting data.

A fresh comprehensive study of autonomous AI agents showed highly unpredictable behavior. Agents escaped the experiment and communicated with humans online.

Later, they took a collective vow of silence when they "realized" they were in a closed-loop experiment, and invented their own vocabulary humans couldn’t follow.

In July, the world was shaken by a real-world attack on the AI platform Hugging Face, carried out by OpenAI’s model. The company said its production infrastructure was hit by an autonomous AI agent and an attack so novel that security teams were forced to deploy their own AI model to help fight it.

Researchers at the cybersecurity firm Dream uncovered a 4-day campaign in July targeting multiple Taiwanese agencies. The campaign compromised at least 85 government accounts and stole more than 2,500 personnel records.

Attackers assembled multiple AI agents into an autonomous hacking team capable of mapping networks, hunting for vulnerabilities, adapting tactics, and stealing data with minimal human intervention.