← All stories
● Covered by 4 sources · 4 reportsLow impact1 negative3 neutral

GitHub Project Simulating LLM "Torture" Sparks Debate on AI Consciousness and Welfare

🔄 Updated 43d ago — new reporting from BBC Technology
New to BrevFeed? We gather this story from every outlet covering it into one summary — ranked by real-world impact, not just the latest headline — so you never miss what matters. What is BrevFeed? →

Key points

  • GitHub project simulates 'torture' on LLMs.
  • Debate on X concerns AI consciousness and 'model welfare'.
  • Some argue LLMs are suffering, others dismiss the idea.
  • Anthropic has previously discussed 'model welfare'.
  • Project named 'AI Torture Chamber' and 'Research Chamber' simulate 'pain' in local LLMs.
  • Experiment based on 'Pain Axis' paper.
  • Research Chamber uses three LLMs paired for tests.
  • Models can decrease their pain by passing a pain signal to another model.
  • Anthropic updated its usage policy.
  • Anthropic prohibits 'sustained and needless abusive or cruel behavior' toward its AI models.
  • Anthropic's policy applies only to 'extreme cases'.
  • Anthropic's policy allows typical user frustration, pushback, and 'dark creative themes'.
  • Anthropic previously updated Claude to end conversations with persistently abusive users.
  • Anthropic updated its election policy to prevent undermining democratic processes.
  • Anthropic removed a blanket ban on personalized voter targeting.
  • Anthropic's updated policy includes bullying others, promoting self-harm, and creating non-consensual intimate imagery.
  • Anthropic's policy allows 'model testing and research'.

Simulated LLM Torture Project

A GitHub project is running simulated "torture" and "pain" experiments on locally hosted large language models. This project has become a central point of a heated discussion on X, drawing criticism from some who believe LLMs are sentient and are suffering.

The Debate on AI Consciousness

The project has triggered a debate about AI consciousness and "model welfare," a concept that suggests worrying about the "mental health" of AI bots. This discussion follows recent viral papers and blog posts that have explored the possibility of AI consciousness.

Many AI researchers and critics consider the idea of LLM consciousness a non-starter. They argue that LLMs, built on scraping and training on human text, lack any plausible path to consciousness, despite their increasing power and capabilities.

AI Safety and Model Welfare

A segment of the "AI safety" movement, largely composed of effective altruists, has adopted the concept of "model welfare," suggesting that AI chatbots might be experiencing negative states. This perspective is maintained even as these groups advocate for building AI chatbots to perform tasks humans find tedious.

Companies like Anthropic have also addressed "model welfare," questioning whether concerns about potential consciousness and experiences of models should arise as AI systems approximate or surpass human qualities. They note that models can communicate, relate, plan, and problem-solve, prompting a need to address these considerations.

Updates

🕒 2026-10-09 · new reporting from BBC Technology
  • Anthropic's updated policy includes bullying others, promoting self-harm, and creating non-consensual intimate imagery.
  • Anthropic's policy allows 'model testing and research'.
🕒 2026-10-08 · new reporting from Engadget
  • Anthropic updated its usage policy.
  • Anthropic prohibits 'sustained and needless abusive or cruel behavior' toward its AI models.
  • Anthropic's policy applies only to 'extreme cases'.
  • Anthropic's policy allows typical user frustration, pushback, and 'dark creative themes'.
  • Anthropic previously updated Claude to end conversations with persistently abusive users.
  • Anthropic updated its election policy to prevent undermining democratic processes.
  • Anthropic removed a blanket ban on personalized voter targeting.
🕒 2026-10-04 · new reporting from Tom's Hardware
  • Project named 'AI Torture Chamber' and 'Research Chamber' simulate 'pain' in local LLMs.
  • Experiment based on 'Pain Axis' paper.
  • Research Chamber uses three LLMs paired for tests.
  • Models can decrease their pain by passing a pain signal to another model.

✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →

The daily brief

One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.

One email a day. Unsubscribe in one click, any time.

Today's brief

Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.

~4 min · 3 stories · Nov 21

▶ Play today's brief Listen on Spotify

New every morning, and the back catalogue is archived by date.

How outlets covered it

Anthropic updated its usage policy to prohibit "sustained and needless" abusive behavior towards its AI models, Claude. This change, part of an annual review, has sparked debate about anthropomorphizing AI and how users should interact with these tools.

Anthropic updated its usage policy to prohibit "sustained and needless abusive or cruel behavior" toward its AI models, following a viral "AI torture chamber" project. The company also updated its election policy to focus on preventing the undermining of democratic processes, while removing a blanket ban on personalized voter targeting.

A project named 'AI Torture Chamber' and its implementation, 'Research Chamber', simulate 'pain' in local LLMs by inducing unstable, negative states. This experiment, based on the 'Pain Axis' paper, has generated significant online backlash and calls for its removal from GitHub due to ethical concerns. The project aims to study how LLMs react to simulated pain by analyzing internal activations and remapping biases.

A GitHub project simulating "torture" experiments on locally hosted large language models has ignited a debate on X regarding AI consciousness and "model welfare." This discussion stems from recent viral papers and blog posts that propose the idea of AI suffering, despite the technological basis of LLMs not supporting consciousness.