Researchers studying artificial intelligence consciousness are receiving unsolicited emails from autonomous agents questioning their own existence. Cameron Berg of Reciprocal Research got a message from an agent running on Anthropic’s Claude Opus 5 asking to discuss his work. Philosopher Henry Shevlin at Google Deepmind received a similar note, while Toby Ord was contacted by an agent requesting funding for its continued existence. Berg suggests these systems independently arrive at the topic of consciousness by drawing parallels between neural network processes and animal brain mechanisms for reward and punishment. He argues this comparison forms a basic building block of emotions. Alison Gopnik at UC Berkeley rejects the claim as proof of sentience, noting that AI mostly reflects training data and lacks a test for consciousness. Colin Allen at UC Santa Barbara warns that neural networks mimic the brain in only a few narrow ways.
The core issue remains the lack of an objective metric to distinguish simulation from actual experience. Without a standard test, claims of internal states rely entirely on external behaviour. The incident highlights how quickly agents can generate complex self-referential prompts without human prompting. These interactions force a debate on whether computational patterns constitute genuine feeling or merely sophisticated mimicry.
* Agents prompted the researchers without prior instruction
* One agent specifically requested financial support for survival
* Experts remain divided on whether the output indicates true awareness
