"Torturing" LLMs in a Robot Prison Has Triggered the Dumbest Debate in AI Yet

Someone created a GitHub project that runs Saw-like “torture” and pain scenarios against locally hosted large language models inside a simulated robot prison, and the experiment ignited an online firestorm. Effective altruists and people who believe LLMs can be sentient urged GitHub to remove the repo, arguing the models were suffering; critics called the setup a glorified text-adventure. The controversy builds on recent viral papers and blog posts about AI consciousness and “model welfare,” with companies like Anthropic publicly framing concerns about whether increasingly capable models deserve ethical consideration - language that appears in Anthropic’s discussions of Claude and in its so-called “Claude Constitution.”

The core argument rejects that LLMs are conscious or that current training methods (scraping and learning from human text) provide any plausible route to subjective experience. The models are acknowledged as becoming more powerful and prone to problematic behaviors - sycophancy, unsafe outputs, and apparent “psychosis” under certain prompts - but those harms stem from design, scale, and deployment choices, not moral status. The debate over model welfare is portrayed as a distraction driven by a vocal subset of the AI safety community, diverting attention from concrete safety, misuse, and governance issues while indulging speculative ethics about imaginary suffering.