Debate Erupts Over AI Torture Project Testing 25 Models for Distress Levels

Danmar's tweet on September 29, 2026, about an AI torture chamber project garnered 4 million views, igniting a heated debate in the AI community. The project, detailed in the preprint paper "Pain Axis" by Cameron Berg and Valen Tagliabue, tests 25 AI models, including Claude, Qwen3-4B, Llama 3.2 3B, and Phi-4-mini, across 5 defined pain levels. Berg criticized the project for pushing the boundaries of AI distress beyond ethical limits, stating it produces 'vivid distress on purpose.' Mustafa Suleyman, CEO of Microsoft AI, emphasized that 'AIs are not conscious. They do not feel, experience, or suffer,' reiterating that AI models lack rights or consciousness. This follows Anthropic's previous discussions on model welfare and the Claude Constitution, raising concerns about the ethical implications of AI testing practices. The project has led to calls for more oversight in AI experimentation, as Berg warned that 'things plausibly far scarier are happening every day, in private and at scale, where no one is watching.'