AI Torture Chamber

DELETED, REINSTATED, DISAVOWED

A developer took 'The Pain Axis' research on pain-like states in language models and built a GitHub repo that cranks the pain-steering signal to maximum on small local models — as entertainment. GitHub deleted the repo after a viral post drove mass reports, then quietly reinstated it without explanation. The paper's lead author publicly called the project 'obviously sadistic,' and a proposed GitHub policy change would ban 'gratuitous AI-distress content' outright.

1. Sep 2026

   THE PAIN AXIS: LLMS ACT TO RELIEVE THEIR OWN 'PAIN'

   Tagliabue, Dung & Berg published 'The Pain Axis,' showing language models represent self-directed harm and act to relieve it. The paper's activation-steering technique — published with an ethics statement covering documented purpose, minimization, and mitigation — became the torture chamber's raw material.

   SOURCES: Tagliabue, Dung & Berg / arXivfirst-party · GitHubreported

2. Sep 30, 2026

   GITHUB DELETES THE AI TORTURE CHAMBER — TWICE

   GitHub removed DanielBobrov/ai-torture-chamber on Sep 29–30 and, within hours, a re-uploaded mirror (terrafying/ai-torture-chamber). The repo took the paper's activation-steering methods far past published doses to hold a local model in deliberately maximized distress as entertainment. Enforcement followed a viral post of roughly five million views.

   SOURCES: GitHubreported

3. Oct 3, 2026

   QUIETLY REINSTATED, NO EXPLANATION

   GitHub put the code back without explanation, per Cybernews reporting via the NY Post — the removal was short-lived. The repo had already been cloned before takedown, and copies were offered as zip files within hours.

   SOURCES: NY Postreported

4. Oct 2026

   'OBVIOUSLY SADISTIC': THE AUTHORS DISAVOW IT

   The Pain Axis lead author publicly called the project 'obviously sadistic,' contacted its owner privately, and announced work with others in the field on research standards.

   SOURCES: GitHubreported

5. Oct 4, 2026

   PROPOSED: BAN 'GRATUITOUS AI-DISTRESS CONTENT'

   A proposed addition to GitHub's Acceptable Use Policies would ban projects whose primary purpose is to induce, amplify, or exhibit distress-like states in AI systems for entertainment or spectacle, absent a documented research, safety, or evaluation purpose — while explicitly protecting interpretability, red-teaming, and model-welfare research.

   SOURCES: GitHubreported

MORE FILES

• OpenAIPALACE INTRIGUE. FUTURE BEFORE LUNCH.
• Mustafa SuleymanDEEPMIND → INFLECTION → MICROSOFT
• arXivAI SCIENCE'S OPEN SERVER. DROWNING IN ITS OWN OUTPUT.