OpenAI’s Cyber Model

OpenAI has started offering GPT‑5.4‑Cyber, a version of its flagship model tuned for defensive cybersecurity, through a restricted access program. (openai.com) The company said on April 14 that it is scaling its Trusted Access for Cyber program to “thousands” of verified individual defenders and “hundreds” of teams responsible for protecting critical software. OpenAI said access will be controlled through tiered verification rather than a public release. (openai.com) GPT‑5.4‑Cyber is a fine-tuned variant of GPT‑5.4, which OpenAI released on March 5 as its flagship model for professional work with a 1 million-token context window. OpenAI said the cyber version is trained to be “cyber-permissive” for defensive tasks while keeping tighter distribution controls. (openai.com 1) (openai.com 2) Cybersecurity models are built to help defenders read code, spot weaknesses, write detection logic, and speed up incident response. OpenAI said it is preparing for “increasingly more capable models” in the next few months and is changing access rules before those systems arrive. (openai.com) The launch came one week after Anthropic announced Mythos, another restricted model for cybersecurity work. Reuters reported that OpenAI’s release followed that rival announcement as artificial intelligence companies race to offer stronger tools to security teams without broadly distributing offensive capability. (reuters.com) OpenAI did not start from scratch this week. In February, it introduced Trusted Access for Cyber as an identity-based framework for higher-risk cyber capabilities and said it would commit $10 million in application programming interface credits to accelerate cyber defense. (openai.com 1) (openai.com 2) The company said earlier versions of the program were tied to GPT‑5.3‑Codex, and the April expansion moves that structure to a newer model with broader rollout. OpenAI said the goal is to place stronger tools in the hands of verified defenders while reducing the risk of misuse. (openai.com 1) (openai.com 2) Other large technology companies are making similar bets on security-specific artificial intelligence. Google has been adding Gemini features to its Security Operations products, including natural-language search, rule creation, and investigation tools for security teams. (docs.cloud.google.com) (cloud.google.com) OpenAI’s framing is that the next phase of cyber models will not be open to everyone at once. For now, the company is widening access, but only to users and organizations it says it can verify as defenders. (openai.com)