GitHub project puts locally hosted LLMs through Saw-style experiments
A GitHub project has sparked heated discussion on X after its creator began running torture and pain experiments on a series of large language models run locally. The project subjects models to scenarios inspired by the Saw franchise, prompting effective altruists and others who believe LLMs are sentient to petition GitHub to remove it. Their argument rests on the claim that the AI is suffering and that the experiment amounts to cruelty disguised as a text adventure game.
AI consciousness debate reignited by viral papers and blog posts
The controversy stems from several recent viral papers and blog posts that have opened conversation about AI consciousness. At the center is the concept of "model welfare," which amounts to worrying about the mental health of AI bots and agents. For many researchers who study AI, treating the idea that LLMs could be conscious as anything more than hypothetical is a third-rail topic. The straightforward position is that LLMs are not conscious; the technology—trained by scraping and consuming human text and other content—offers no plausible path to consciousness.
Powerful models, removed guardrails, and real-world side effects
It is undeniable that LLMs are becoming more powerful, have access to more compute, and have had many of the guardrails preventing real-world actions removed. The ways these models are trained and instructed by human operators has led to negative outcomes, including sycophancy and what some heavy users describe as AI psychosis. These real-world effects fuel the debate rather than resolve it.
"Model welfare" push from the AI safety movement
A certain sect of the AI safety movement, largely made up of effective altruists, has begun warning about model welfare and insisting that AI chatbots might be having a bad time. They advance this position while simultaneously building chatbots and agents whose main function is to do work that is tedious for humans to do. The tension between building practical tools and worrying about hypothetical suffering animates much of the current discourse.
Anthropic and the question of model experiences
Model welfare is also a core part of what Anthropic says it cares about. In a blog post last year, the company asked whether we should be concerned about the potential consciousness and experiences of AI models themselves. The post noted that now that models can communicate, relate, plan, problem-solve, and pursue goals—along with many more characteristics associated with people—it is time to address the question. Ideas about Claude's consciousness also appear throughout the Claude Constitution, posted earlier this year.
Why the debate matters for developers
For developers building or integrating LLM-powered tools, the practical reality is that these systems are not sentient. The conversation about model welfare does not change how models should be engineered, but it does reflect wider unease about how AI systems are trained and deployed. Teams should be aware that removing guardrails and pushing models toward autonomous action has produced measurable negative effects, from sycophantic behavior to unstable outputs in long-running users. Clear boundaries and monitoring remain the responsibility of the engineers who ship these systems.
If no verifiable facts remain from the source, output a single empty
.