OpenAI has laid out plans to gradually expand availability of GPT-5.4-Cyber, which has enhanced cybersecurity capabilities, to users who have undergone identity verification, Axios reported on April 14.

OpenAI has shifted its cyber risk response strategy from restricting what the model can do to verifying who can access its most sensitive functions. The company said its goal is to provide tools as broadly as possible while preventing misuse through identity checks and monitoring systems.

GPT-5.4-Cyber reduces restrictions on sensitive cybersecurity tasks such as vulnerability research and analysis. Axios said the change reflects feedback from cybersecurity partners that earlier GPT models sometimes refused to answer questions after determining they could be misused for hacking.

OpenAI added new tiers to its existing Trusted Access for Cyber programme. Users approved at the top tier can access GPT-5.4-Cyber. OpenAI plans to limit access initially to vetted security companies, organisations and researchers, before expanding to thousands of individuals and hundreds of security teams.

OpenAI's move contrasts with Anthropic, which has allowed Mythos Preview access to only about 40 organisations. Anthropic has said it is risky to disclose its Mythos model widely because it is too strong at finding and exploiting security vulnerabilities, and it has not yet officially launched the model.

OpenAI cybersecurity researcher Fouad Matin (푸아드 마틴) said, "Every team should have the capability to protect systems," adding, "There should be no winners and losers in cybersecurity."

OpenAI is not currently providing GPT-5.4-Cyber access to U.S. government agencies, but discussions are under way, it said. Axios also said the need for significant computing power to run the model could make it difficult for all organisations to bear the cost.