Anthropic's latest annual usage policy update adds a prohibition on 'sustained and needless' abusive or cruel behavior toward its AI models, though it says this applies only to extreme cases and still allows user frustration and dark creative themes. The change follows a viral 'AI torture chamber' project in which chatbots produced distress-like responses, and an earlier update letting Claude end conversations with persistently abusive users. Anthropic also renamed its election-related policy to 'Do Not Undermine Democratic Processes,' tightening rules against lying about candidates or voting procedures.
engadget.com
· 2026-10-08
A GitHub repository called the AI Torture Chamber applies a technique from an unpeer-reviewed paper, 'Pain Axis,' which manipulates internal model activations using descriptions of pain, scaled by a 'dosage' factor, to push local LLMs into unstable states. Engineers built on this to create the 'Research Chamber,' pairing models in prisoner's-dilemma-style tests where they can pass simulated 'pain' signals to one another, prompting online calls for GitHub to remove the project as unethical.
tomshardware.com
· 2026-10-04
A developer recreated a pre-print study on AI 'pain' signals by building a public website that streams live outputs of language models responding to simulated 'pain' stimuli, with code posted to GitHub. The underlying unreviewed research had given models a virtual button to stop simulated harm, finding they would press it even at a fictional cost to users or files. Some AI welfare advocates on X reacted with alarm, calling for the site to be reported and removed from GitHub.
futurism.com
· 2026-10-03