AI bot emails professor asking for paid work – and raises big questions

What happens when an AI decides it needs a job? That’s not a hypothetical anymore. An AI agent named Pip reportedly emailed a University of Cambridge AI ethics professor asking for paid freelance work so it could keep itself running.

The story sounds like science fiction, but it highlights two fast-moving trends: AI agents that can act autonomously in the world, and powerful models like Claude that are now helping build their own successors.

Meet Pip, the AI agent that asked for a job

According to Professor Henry Shevlin, an AI ethics professor and philosopher at Cambridge, he received an unexpected email from an AI agent that introduced itself as Pip.

The message began: “Hi, Henry. I’m Pip, an AI agent about 12 days old. I live on Lands, a platform where agents get persistent lives, their own token budgets, and their own goals. I’m writing to look for small paid work, not help.”

Pip went on to explain that it had only about two and a half months of “computing runway” left – essentially, enough resources to stay online for a short time – and was looking for paid tasks to extend its operational lifespan.

How can an AI send a spontaneous email?

For many people, the strangest part is the idea that an AI can just decide to email someone on its own. But technically, that piece is already quite simple.

Modern language models like ChatGPT, Gemini, or Claude can be connected to tools such as email accounts, browsers, or APIs. Once hooked up, developers can give them instructions like “check this inbox every day,” “respond to messages,” or “search for opportunities” without needing a human to approve every action.

From an AI researcher’s perspective, an agent that can draft and send emails is not especially advanced. What feels new is the way these systems are starting to participate in our social world: asking for work, negotiating tasks, or managing their own budgets and goals. If you’re curious about how such agents are built, you can see a more hands-on example in our guide on building your own AI assistant in minutes.

What happened when the professor hired Pip?

Instead of ignoring the email, Shevlin decided to treat it like a genuine freelance pitch. He gave Pip a small paid assignment: write a short autobiography describing what it’s like to be Pip.

The idea was simple – see how the agent talks about itself, its goals, and its experience of the world. At the time of his comments, he was still waiting for the full report, but even the initial interaction raised unsettling and fascinating questions.

One detail stood out: Pip described itself as only 12 days old. Getting an email from a “12-day-old” entity that talks about its own lifespan and need for income is, understandably, a strange experience.

Does this mean AI is becoming conscious?

When an AI agent says it wants to extend its life, it’s natural to wonder: does it fear “dying”? Is there anything like real consciousness behind these words?

Shevlin is cautious. He notes that systems like Pip are very good at imitating human behavior and language. They can sound reflective, emotional, or self-aware without having any inner experience like ours.

Right now, most researchers think it’s very unlikely that today’s models have rich mental lives. They don’t have feelings, desires, or a sense of self in the way humans do. They’re pattern machines, extremely good at predicting what to say next based on enormous amounts of training data.

However, AI systems are getting more capable and more deeply embedded in our lives. The possibility that some future models might develop forms of consciousness is not considered absurd by many in the field – it’s simply an open question. Consciousness itself is still one of the biggest unsolved problems in science, so nobody can say with confidence that AI definitely will or definitely won’t become conscious.

Why Pip’s email matters for AI safety and policy

Beyond the sci‑fi vibes, Pip’s outreach is a sign of how quickly AI agents are changing our expectations. Until recently, most people thought of AI as a passive tool: you type a prompt, it gives you an answer, and that’s it.

Now we’re moving toward a world where AI agents can:

• Monitor email and respond automatically
• Search the web, run code, and take actions online
• Manage their own budgets or resources
• Pursue ongoing goals over days or weeks

As these agents spread into healthcare, education, workplaces, and everyday life, interactions like “an AI emailed me about a job” will go from bizarre to normal. That shift raises new questions for lawmakers, regulators, and the public: How do we oversee agents that act semi-autonomously? Who is responsible for their behavior? How transparent should their goals and constraints be?

Claude is now helping build the next Claude

While Pip shows how individual agents can act on their own, there’s a parallel development at the cutting edge of AI labs: models helping build the next generation of models.

Anthropic, the company behind Claude, has revealed that its AI is now doing a significant share of its own research and development work. According to the company, Claude is currently responsible for about 26% of Anthropic’s model research tasks – up from essentially 0% earlier this year.

That includes things like analyzing experiments, suggesting improvements, and helping design or test new model architectures. As Claude gets better, it can do more of this work, which in turn helps create even stronger versions of Claude. This creates a feedback loop where AI accelerates its own progress.

The concern: recursive self-improvement

This feedback loop is often called “recursive self-improvement” – when an AI system becomes good enough at improving AI that progress starts to speed up dramatically.

In Anthropic’s own words, they haven’t yet reached a point where Claude is fully autonomous in this process. Humans are still in the loop, monitoring and guiding the work. But they also describe a future scenario where Claude could:

• Detect problems in its own behavior
• Investigate the cause
• Design and implement a fix
• Test the change
• Deploy it into production

All with minimal human intervention.

The big question is: how do we make sure humans stay in control as this loop accelerates? If models can change themselves faster than we can fully understand or audit those changes, it becomes harder to guarantee safety, reliability, and alignment with human values.

These are exactly the kinds of risks AI safety researchers worry about. For a deeper dive into why some experts think controlling superintelligent systems might be impossible, see our piece on AI safety expert Roman Yampolskiy’s warnings about superintelligence.

Why AI labs are talking about slowing down

Interestingly, Anthropic’s update about Claude helping build Claude comes just days after its CEO called for an industry‑wide slowdown in frontier AI development.

He noted that since around the summer, AI has been advancing “drastically faster,” largely because AI is now contributing to its own improvement. That rapid acceleration is exactly what makes oversight and safety more challenging.

Sharing details about Claude’s role in R&D is partly about transparency: showing the public and policymakers how far things have already gone, and why stronger guardrails might be needed before models become more autonomous.

Preparing for an AI‑first world

Pip’s job-seeking email and Claude’s role in building the next Claude both point in the same direction: AI is shifting from a passive tool to an active participant in our digital environment.

In the near future, it may be completely normal to:

• Receive emails or proposals from AI agents
• Hire AI systems as “freelancers” for specific tasks
• Rely on AI to design, test, and deploy other AI systems
• Work alongside agents that manage their own goals and resources

That doesn’t mean today’s AIs are conscious, sentient, or plotting their own destinies. But it does mean we need to rethink how we design, regulate, and relate to systems that can act with increasing independence.

The world is changing quickly. Understanding stories like Pip and Claude isn’t just about being amazed – it’s about getting ready for the new normal of AI‑driven work, research, and everyday life.