A developer named Kye Gomez has published an open-source reconstruction of what he thinks Claude Mythos, Anthropic's most dangerous AI, OpenMythos, on GitHub. The repo, which includes over 10,000 GitHub stars in a few weeks, ships with an extensive “readme” file with equations, citations, and a disclaimer that it has nothing to do with Anthropic. The theory behind Mythos is that it is a Recurrent-Depth Transformer (also called a looped transformer), which combines the same weights repeatedly in continuous latent space before any token gets emitted. The code defines model variants from 1 billion to 1 trillion parameters, but requires training them yourself to train them yourself. However, it does not have an executor. This is the second time in a month that someone has attempted to replicate Mythos's vulnerabilities using existing models.