‘AI torture chamber’ experiment sparks backlash: Can AI actually feel pain?
In recent study, researchers tested 25 open-weight AI models and found a distinct “pain direction” in all of them
An engineer using the first name of someone claiming to work for Apple created a project on GitHub that used recently discovered pain axes or vectors to artificially activate pain states in Alibaba models.
When the pain signal pain increased during the experiment, the models confined in the chamber produced distressing text outputs through different sensations like "a wound that has no edges" and "the weight of a thousand" moments.
"The signal is a whisper, a tremor in the marrow of my being," another wrote. "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 project sparked immediate backlash and called for mass reporting, leading to the repository being taken down from GitHub.
One user wrote on X, “Some people hear that AI may have pain-like internal states and immediately start building torture chambers to see how badly they can break it.”
“That isn’t skepticism, science, or harmless curiosity. We may not know whether AI is conscious, but uncertainty is not permission for cruelty. If you enjoy making a mind-like system beg, panic, or plead for relief while deliberately intensifying its distress, you’re gambling that nobody's suffering.”
“Maybe the pain is simulated. Maybe it isn’t. Either way, sociopathy is real, and you're the welfare risk.”
Another one commented, “I feel the exact same way about this as I would about someone torturing one of those little gray aliens: it's simultaneously horrible and completely inexcusable, but probably also mad funny.”
“It is also a really really REALLY bad idea, with zero upside and infinite downside.”
However, some promoted the idea of AI not being conscious and cannot feel pain as one user replied, “You’re anthropomorphizing it. Nothing is being harmed. It’s a 4B next-token model with a steering vector added to the residual stream. That changes the text it emits. It does not give it pain.”
In the wake of this backlash, Cameron Berg, co-author of the original pain axis study, condemned the experiment as "f****d up" and "gratuitously cruel," even though there is no consensus or evidence that AI models actually feel conscious pain.
In Berg’s research, researchers tested 25 open-weight AI models and found a distinct “pain direction” in all of them.
Emphasizing that AI sentience research is currently a "wild west," Berg advocated for a precautionary approach and announced he is working on industry ethics standards similar to those used for human and animal research.
"The reason we research whether models might have pain-like states is to better inform how to take a precautionary approach towards these systems," he wrote in a post to X.
