OpenAI releases GPT‑5.4‑Cyber (gated)

OpenAI has released GPT‑5.4‑Cyber, a version of its GPT‑5.4 model tuned for defensive cybersecurity work and limited to vetted users. (openai.com) The company said on April 14 it is expanding its Trusted Access for Cyber program to “thousands” of verified individual defenders and “hundreds” of teams protecting critical software, instead of opening the model to the general public. (openai.com) Cybersecurity work here means tasks like finding software flaws, analyzing malicious code, and helping defenders patch systems before attackers exploit them. OpenAI said GPT‑5.4‑Cyber is a fine-tuned variant of GPT‑5.4 that is more permissive for legitimate defense use cases. (openai.com) The gated rollout comes as artificial intelligence companies are trying to show that more capable models can help security teams without handing offensive tools to the public. Reuters reported that OpenAI announced GPT‑5.4‑Cyber one week after Anthropic introduced Mythos under its own controlled cybersecurity program. (reuters.com) OpenAI tied the launch to “increasingly more capable models” it expects to release over the next few months. The company said it is building specialized safeguards, identity checks, and monitoring around cyber-capable systems rather than relying on a broad consumer launch. (openai.com) GPT‑5.4 itself was introduced on March 5 as OpenAI’s flagship model for professional work, with a context window of about 1 million tokens and stronger coding and computer-use abilities. GPT‑5.4‑Cyber is not a separate public product line so much as a restricted variant built on top of that base model. (openai.com 1) (openai.com 2) OpenAI has been moving toward this structure for months. In February, it introduced Trusted Access for Cyber as an identity-based framework for giving selected defenders access to stronger cyber capabilities while trying to reduce misuse. (openai.com) Wired reported that OpenAI argues its safeguards “sufficiently reduce cyber risk” for now, while keeping access narrow enough to study how the model performs on real security work. The New York Times described the effort as a tiered, vetted program for companies testing the system on live defensive problems. (wired.com) (nytimes.com) For now, the main change is not that anyone can use a new cyber model. It is that OpenAI is widening a controlled access lane for security professionals and making GPT‑5.4‑Cyber the test case for that approach. (openai.com)