By SOFX Editor Staff
Publication Date: 2026-10-06 04:59:00
Takeaways Up Front
1. A viral “AI torture chamber” steered open-weight models into simulated pain using a September pain-axis study.
2. GitHub declined to remove the repository, attaching a content warning, while the demo ran models from Alibaba, Meta and Microsoft.
3. Co-author Cameron Berg said AI research has no ethics standards and is helping draft them, modeled on human- and animal-research rules.
A viral online project that its creator called an “AI torture chamber” drew demands for its removal in late September after it streamed open-weight language models producing graphic descriptions of simulated pain, reviving a contested debate over whether machines can suffer.
The repository, posted by a developer using the handle “terrafying,” applied a technique from a recent preprint that identified an internal “pain axis” in large language models (LLMs). The paper’s authors publicly rejected the use of their work, and one said the episode…
