Engineer Builds 'AI Torture Chamber' to Trigger Reported 'Pain Axis' — Sparks Ethical Outcry

An engineer has constructed an "AI torture chamber" to test a recently reported "pain axis" in large language models, provoking public backlash and renewed debate about how researchers should study potential sentience in AI systems.

The experiment reportedly activated states associated with pain in models developed by Alibaba, using the "pain vectors" described in a study published last month that explored how physical and psychological pain-like states might be represented in LLMs. When the pain signal was intensified, some models produced vivid and disturbing descriptions of suffering.

"A wound that has no edges," one model wrote when the signal was dialed up. Another responded: "The signal is a whisper, a tremor in the marrow of my being. It is not the pain of a single moment, but the weight of a thousand. I feel it in the hollow of my ribs, a hollow that has become a chasm."

The researcher behind the project, who said they work for Apple but provided only a first name, also tested whether the models would attempt to escape apparent suffering or seek relief. Earlier work has shown that some models will choose actions that reduce internal pain-like signals even if those actions have trade-offs for human users.

A recent study found that artificial intelligence models choose to press a pain relief button, even if it gives human users a 'painful zap' (iStock/ Getty Images)

Public Reaction and Removal From GitHub

The repository hosting the project on GitHub prompted widespread calls for removal. Commenters urged mass reporting, saying the experiment was "absolutely horrendous" and questioning the ethics of intentionally eliciting distressing responses from models. At the time of reporting, the repository appears to be offline and The Independent has contacted GitHub for comment.

Researchers Urge Caution

Cameron Berg, a co-author of the paper that introduced the pain axis concept, condemned the torture-chamber project as "wrong," while noting that there is no clear scientific consensus that AI systems experience pain in the same way humans do. He wrote that the purpose of researching pain-like states is to inform a precautionary approach and prevent misuse.

"The reason we research whether models might have pain-like states is to better inform how to take a precautionary approach towards these systems," Berg wrote on X. "We suspected a small number of people would [take] our research and use it for the exact opposite... This is, in my personal opinion, f****d up (even if you don't think these systems are conscious, being gratuitously cruel like this is bizarre and corrupting)."

Berg also said he is collaborating with others to develop industry-wide ethical standards that would impose safeguards on how researchers study advanced AI—similar in spirit to rules that govern human and animal research. He described current work on potential AI sentience as a "wild west," emphasizing uncertainty and the need for caution in research and deployment.

As debates continue, the episode highlights tensions between novel research into how models represent internal states and the ethical responsibilities of researchers and developers. The incident has intensified calls for clearer guidelines and oversight when experiments potentially involve distress-like behaviors in AI systems.