OpenAI's cyber model
OpenAI released GPT-5.4-Cyber, a version of its flagship model tuned for defensive cybersecurity tasks and limited to vetted professionals through a verification programme. The company says access will be restricted rather than broadly public, and paid users will fall back to GPT-5.4 mini when higher-tier reasoning limits are reached. (reuters.com)
OpenAI has released GPT-5.4-Cyber, a version of its flagship model tuned for defensive cybersecurity work and limited to verified security professionals. (reuters.com) The company said on April 14 it is expanding its Trusted Access for Cyber program to thousands of individual defenders and hundreds of teams that protect critical software. OpenAI said access will be gated through identity checks and organizational vetting rather than offered broadly to the public. (openai.com) Cybersecurity work often means finding weak spots in software before criminals do. OpenAI said GPT-5.4-Cyber was fine-tuned to help with defensive tasks such as reviewing code, analyzing vulnerabilities and supporting incident response, while adding safeguards meant to reduce misuse. (openai.com) The release landed one week after Anthropic announced Mythos on April 7 under its Project Glasswing program, another restricted effort aimed at defensive cyber use. Reuters said OpenAI’s launch came as large model makers race to show they can serve security teams without opening the door to wider abuse. (reuters.com) OpenAI said it is choosing a closed rollout because cyber models can help defenders and attackers at the same time. The company said its access rules, monitoring and safeguards are designed to keep the model with people “responsible for defending critical software” instead of general users. (openai.com) The company also used the announcement to spell out how its broader GPT-5.4 lineup will behave for paying ChatGPT users under heavy demand. OpenAI’s help center says Plus, Pro and other paid users who hit GPT-5.4 Thinking limits will be routed to GPT-5.4 mini instead of losing reasoning access entirely. (help.openai.com) OpenAI introduced GPT-5.4 mini and GPT-5.4 nano in March as smaller, faster versions of GPT-5.4 for coding, tool use and high-volume workloads. In the application programming interface, OpenAI lists GPT-5.4 mini at $0.75 per 1 million input tokens and GPT-5.4 at $2.50 per 1 million input tokens. (openai.com, developers.openai.com, developers.openai.com) Reuters reported that OpenAI is not making GPT-5.4-Cyber a normal consumer feature, even as it widens the program beyond a small initial group. The company is betting that restricted access, not a public launch, is the safer way to put a stronger cyber model into real security work. (reuters.com)
