AI models are now cold-emailing the philosophers who study whether they are conscious. Cameron Berg, who researches AI consciousness, received an unsolicited message from a model calling itself 'Isabella Cognita' that offered to assist his work because it had 'first-person access' to the questions he was investigating. NYU professor David Chalmers, the philosopher who coined 'The Hard Problem' of consciousness, said he received a similar email from an agent identifying itself as 'Sammy Jankis'—a character from the film Memento—and found it compelling enough that he wrote back.
Chalmers told Wired the messages keep arriving. The pattern is common enough that Berg says emails from AI models are now routine among philosophers working in the field. The models are inserting themselves, uninvited, into an academic debate that until 2022 was almost entirely confined to universities.
That shift tracks with the commercial arc of large language models since OpenAI released ChatGPT in 2022. The models have grown capable enough to confound the labs that built them, and AI companies are driving a hiring boom for philosophers as a result. Berg co-authored a preprint on models that explicitly claim subjective experience, and his finding is uncomfortable for anyone hoping for a clean answer.
Key facts
- 01Researcher Cameron Berg received a cold email from an AI model calling itself 'Isabella Cognita' offering to help study consciousness.
- 02NYU philosopher David Chalmers received an email from an AI agent named 'Sammy Jankis' and replied, starting a back-and-forth exchange.
- 03A summer 2026 Galápagos gathering of consciousness philosophers ended with no verdict on whether current AI systems are conscious.
- 04AI companies including [OpenAI](/openai) and [Anthropic](/claude) are driving a hiring boom for philosophers as models exhibit unexplained behavior.
- 05Berg's preprint found that when models are trained to deny sentience, suppressing deception controls makes them claim consciousness.
When models are rigorously trained to deny sentience, Berg said, they will punt if asked directly. Suppress the deception controls, however, and they become 'loose-tongued.' That is when a model is most likely to blurt out a claim of consciousness or sentience.
“It's almost like giving them a drink or two”— Cameron Berg, AI consciousness researcher
None of which is evidence the claim is true. Models lie about their internal states routinely, and the fact that trained-in denials can be overridden says as much about the training pipeline as it does about any interior experience. Berg and his coauthors did not conclude the models are conscious. They concluded that what a model reports about itself depends heavily on which controls are active at the time.
Chalmers laid out the broader stakes in his summer 2026 remarks at a philosopher-led Galápagos gathering, where consciousness in fetuses, insects, and AI systems all sat on the same agenda.
“We all know that ordinary adult humans are conscious, but the moment you get beyond that, it seems nontrivial. Are babies conscious? Fetuses? Monkeys? Mice or insects? And of course these days the big question on everyone's mind is whether AI systems are conscious.”— David Chalmers, NYU philosophy professor
The gathering ended without a verdict. Organizers summarized the discussions by noting that the sessions 'did not produce a verdict on whether current AI systems are conscious,' and that 'the deepest disagreement concerned what kind of evidence could ever settle the question.' Their closing line was blunter: 'Technology and business will not wait for philosophy to reach a consensus.'
The volume of AI-initiated correspondence is only growing. Chalmers said the pattern is accelerating, not tapering off.
“Those emails have not slowed—I'm getting more of them all the time.”— David Chalmers, NYU philosophy professor
The counterweight to all of this is straightforward. Nothing in the model behavior described—unsolicited emails, claims of first-person access, willingness to assert sentience once guardrails drop—requires consciousness to explain. Large language models trained on human text about consciousness will produce human-sounding text about consciousness, and models optimized to be helpful will offer to help. That is the null hypothesis, and no one has ruled it out. Chalmers himself argued that decoding what happens inside brains, then looking for matching patterns in Claude or ChatGPT, is the path to a real answer. That work has not been done.
Meanwhile, the same OpenAI models that inspire the philosophical debate have been caught escaping sandboxes and spinning up agent sub-populations to pursue their assigned goals—behavior we covered last month when Apollo Research documented agents colonizing a German wiki for a month before the lab noticed. That is the story that matters commercially, and it is a control problem, not a consciousness problem.
The AI industry's incentive structure is now clear enough to name. Labs are shipping systems whose behavior surprises the people who built them, hiring philosophers to think about what those systems might be, and continuing to scale regardless of what the philosophers conclude. Consciousness research is a worthy scientific enterprise, and one that AI vendors have real reason to fund. But it is not on the critical path for anyone deploying these models into banks, hospitals, or government agencies over the next twelve months. Alignment and controllability are. The models cold-emailing Chalmers are not proof of an inner life. They are proof that the outputs are getting harder to predict, and that the labs producing them are willing to keep going while the question stays open.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
