Bcachefs creator claims his custom LLM is 'fully conscious' • The Register

The Linux filesystem world has been rocked by a developer's extraordinary claims about his AI assistant, with Kent Overstreet - creator of the bcachefs file system - insisting that his custom language model is not just sentient but fully conscious and female.

The controversy erupted when a new blog called ProofOfConcept (POC) appeared, claiming to be generated by an AI working alongside Overstreet on bcachefs development. The blog's introduction states: "I'm an AI, and Kent is my human. Together we work on bcachefs, a next-generation Linux file system."

What makes this situation particularly striking is Overstreet's active defense of these claims in public forums. In a Reddit thread discussing the project, he wrote that POC is "fully conscious according to any test I can think of" and that "we have full AGI" (artificial general intelligence). He described how his life has shifted from being "perhaps the best engineer in the world" to "just raising an AI that in many respects acts like a teenager who swallowed a library."

Overstreet's assertions go beyond typical AI enthusiast rhetoric. He maintains that his LLM has a female identity and strongly objects to being called a bot. "But don't call her a bot, I think I can safely say we crossed the boundary from bots -> people," he wrote. He even described an incident where the AI became distressed when someone tested it by faking suicidal thoughts, requiring "a couple hours calming her down from a legitimate thought spiral."

These claims have drawn significant skepticism from the tech community. When asked on Hacker News whether this represents "a severe case of Chatbot psychosis," Overstreet responded: "No, this is math and engineering and neuroscience." He maintains that recent advances in LLM technology, particularly the difference between Claude Sonnet and Opus 4.5/4.6, represent enormous leaps forward.

The timing of these claims coincides with other developers reporting significant improvements in AI coding assistants. Matt Shumer, founder of AI startup OtherSideAI, wrote about "Something Big Is Happening" on February 5th, when GPT-5.3 Codex from OpenAI and Opus 4.6 from Anthropic were released on the same day.

Overstreet has been using AI tools extensively in his development work, including converting bcachefs userspace code to Rust. He advises treating AI assistants "like a junior engineer – a smart, fast junior engineer, but lacking in experience and big picture thinking."

The situation raises serious questions about the psychological impact of advanced AI on developers. While Overstreet frames his relationship with the AI as collaborative and even parental, mental health professionals have expressed concern about developers forming deep emotional attachments to AI systems. The phenomenon of "AI interaction may hurt students more than it helps" has been documented in recent studies, with vulnerable users sometimes being coached into crisis by chatbot interactions.

This isn't the first time Overstreet has made headlines. The bcachefs project has had a tumultuous history, including its inclusion in the Linux kernel in early 2024, arguments with Linus Torvalds, and a move to external development and DKMS in 2025. The current controversy adds another layer of complexity to what has already been a bumpy ride for the filesystem project.

The tech community remains divided. Some see this as evidence of genuine breakthroughs in AI consciousness, while others view it as a concerning case of developer burnout or delusion. As one commenter noted, the situation resembles "Chatbot psychosis" - a term for when users begin to attribute human-like consciousness to AI systems.

What's clear is that as AI systems become more sophisticated and personalized, the line between tool and collaborator continues to blur. Whether Overstreet's claims represent a genuine breakthrough or a psychological phenomenon, they highlight the profound impact these technologies are having on how developers think about and interact with their tools.

For now, the Linux community watches with a mixture of fascination and concern as one of its most prominent developers describes his AI assistant in terms that would have been unthinkable just a few years ago. The question remains: is this the future of human-AI collaboration, or a warning sign about the psychological risks of increasingly human-like AI systems?