OpenAI’s GPT-5.4-Cyber Challenges the “Superhuman” Hacking of Claude Mythos

Anthropic and OpenAI have introduced specialized AI models for cybersecurity analysis: Claude Mythos (restricted to selected enterprises) and GPT-5.4-Cyber (designed for defensive use with an application process that allows vetted individual researchers and organizations). Both products are tightly access-controlled, require authorization or identity verification, and are explicitly framed for defensive research rather than offensive use.