OpenAI works on cybersecurity model to rival Anthropic Mythos

Anthropic pointed out that Mythos Preview did not get any special security training. Its ability to find vulnerabilities comes from general improvements in code understanding and logical reasoning, not from targeted preparation. This has both benefits and risks, because the same skills that help it fix flaws could also let it exploit them.

Claude Mythos Preview first surfaced publicly after internal draft materials were left in an unprotected, publicly accessible data store on Anthropic's servers. The company attributed the exposure to a configuration error in an external content management tool. Those documents identified the model under the internal code name "Capybara" and indicated it had beaten Claude Opus 4.6 on benchmarks spanning cybersecurity, coding, and academic reasoning; they also described the system as occupying a position in Anthropic's hierarchy that would sit above the Opus line.