OpenAI Launches GPT-5.4-Cyber to Challenge Anthropic's Mythos in the AI Cybersecurity Arms Race

The AI-powered cybersecurity race has reached a new level. OpenAI has announced GPT-5.4-Cyber, a modified version of its GPT-5.4 model built specifically for cybersecurity defense operations. Instead of restricting what its models can do, OpenAI is now prioritizing rigorous identity verification to determine who gets access to its most powerful cybersecurity capabilities. The model features lower refusal boundaries for legitimate cybersecurity tasks — what OpenAI calls being "cyber-permissive" — and includes binary reverse engineering capabilities, allowing analysts to hunt for malicious code in programs and apps. The model is being offered to participants in OpenAI's Trusted Access for Cyber program, launched in February. OpenAI plans to expand access to thousands of individuals and hundreds of security teams, provided they complete verification checks. Rival Anthropic's Mythos is considered so powerful it was shared with only a handful of the biggest software companies — including Microsoft, Apple, and Google — after the tool reportedly found decades-old vulnerabilities with ease. OpenAI says the GPT-5.4-Cyber release is also preparation for an even more capable AI model due later this year.

‍