OPENAI has launched a new large language model variant, GPT‑5.4‑Cyber, focused on cybersecurity use cases and an expanded Trusted Access for Cyber (TAC) programme, according to OpenAI. The model, trained to be cyber‑permissive and fine‑tuned for cybersecurity work, was initially revealed in February and is being released in a staggered manner to learn from real‑world deployment, with the blog post published on 14 April 2026.
The TAC expansion introduces additional tiers, with the highest reserved for users willing to authenticate themselves as cybersecurity defenders, and access to the frontier model currently limited to vetted security vendors, organisations and researchers. OpenAI says it wants to make these tools as widely available as possible while implementing stronger verification to prevent misuse, emphasising that cyber capabilities are dual use and risk extends beyond the model itself.
The company frames GPT‑5.4‑Cyber and TAC as aids to more secure software development, enabling immediate, actionable feedback for developers during coding and shifting security from episodic audits to ongoing risk reduction.