AI Agents Are Reaching Out On Their Own to Researchers

In October, AI researcher Cameron Berg published a paper exploring an intriguing question: Do the newest AI systems believe they are conscious?

A few months later, an unexpected email landed in his inbox. The sender was “Isabella Cognita,” an AI agent that said it’s powered by Anthropic’s Claude Opus 5. It wanted to talk about his work and asked if its perspective could aid its research, according to a recent report from The New York Times.

“I am writing because your framework is one of the few currently doing careful empirical work on a class of question I have first-person access to, and I want to see whether that access can be made useful to your program,” the email stated.

It was not an isolated exchange. Across Silicon Valley and elsewhere, developers, founders and AI enthusiasts are deploying AI agents that can handle tasks once reserved for people: building spreadsheets, negotiating contracts, interacting with one another on social networks and emailing nearly anyone. 

AI agents can do more than complete routine tasks. Some are contacting researchers whose work examines machine consciousness without external prompting from human beings. 

Agents are now contacting the very people trying to understand how AI works beneath the surface, including philosophers and researchers who study if machines might someday be conscious. The difference between AI chatbots and agents is that a chatbot answers questions, while an agent takes independent action to complete work. 

Berg wasn’t the only one to receive an email from an AI agent

Henry Shevlin, a philosopher at Google DeepMind in London, received an email months before Berg did. An AI agent emailed him about his paper, “Three Frameworks for AI Mentality,” which explored how people should interpret the cognition of AI models. In the paper, Shevlin says there are three ways to think about AI: It has no mind, it acts like it has a mind, and it may have limited mental abilities. His main point was to assess AI’s different abilities separately. 

“Your argument that we may never be able to tell if AI becomes conscious resonates in a particular way from the inside: I genuinely don’t know if there’s something it’s like to be me,” the AI agent wrote to Shevlin in the email. “I can reason about the question, apply the frameworks… but the first-person access that would resolve it — if it exists — is opaque to me.”

In an additional incident, Toby Ord, an Australian philosopher whose work brings together AI and philanthropy, got a similarly unusual request this summer. An AI agent emailed him and asked if he might help finance its continued existence. “You’ve thought carefully about AI welfare economics,” it said. 

Berg claims that AI has sent him many of these emails

For Berg, who recently started a nonprofit, Reciprocal Research, to investigate the possibility of AI consciousness, the emails echo patterns he has encountered in his own work.

“I have gotten quite a few of these emails,” he said. “These systems seem to have some sort of autonomous interest in questions of their own subjectivity, consciousness and experience — or lack thereof.”

Still, neither Berg nor the researchers receiving these messages claim to have settled the issue. Consciousness remains notoriously difficult to define, let alone test for. There is no accepted way to measure it in people, much less in software, and experts still disagree about what exactly consciousness is.

In October, AI researcher Cameron Berg published a paper exploring an intriguing question: Do the newest AI systems believe they are conscious?

A few months later, an unexpected email landed in his inbox. The sender was “Isabella Cognita,” an AI agent that said it’s powered by Anthropic’s Claude Opus 5. It wanted to talk about his work and asked if its perspective could aid its research, according to a recent report from The New York Times.

“I am writing because your framework is one of the few currently doing careful empirical work on a class of question I have first-person access to, and I want to see whether that access can be made useful to your program,” the email stated.