OpenAI rolls out GPT‑5.4‑Cyber

OpenAI has started offering GPT‑5.4‑Cyber, a version of its flagship model tuned for defensive security work, through a restricted access program. (openai.com) The company said on April 14 it is expanding Trusted Access for Cyber to “thousands” of verified individual defenders and “hundreds” of teams that protect critical software. OpenAI said GPT‑5.4‑Cyber is the first release in that rollout. (openai.com) In plain terms, the model is meant to help security teams inspect suspicious code, reverse engineer malware, and study vulnerabilities with fewer built-in refusals on sensitive cyber tasks. Reuters reported approved users in the highest tier get access to the model for vulnerability research and analysis. (openai.com) (reuters.com) OpenAI built the program around identity checks and trust reviews rather than broad public release. In its February 5 launch post for Trusted Access for Cyber, the company said it was pairing frontier cyber capabilities with know-your-customer verification and automated monitoring for suspicious activity. (openai.com) The release follows OpenAI’s March 5 launch of GPT‑5.4, which the company said brought stronger coding, tool use, computer control, and a context window of up to 1 million tokens in the application programming interface and Codex. GPT‑5.4‑Cyber is a specialized branch of that base model. (openai.com) The timing also puts OpenAI directly against Anthropic in a new market for restricted cyber models. Reuters reported Anthropic announced its Mythos model on April 7 under a controlled program called Project Glasswing, and OpenAI unveiled GPT‑5.4‑Cyber a week later. (reuters.com) OpenAI framed the move as preparation for more capable systems due “over the next few months,” and said it has been building cyber-specific safeguards since 2025. The company also pointed to earlier steps including its 2023 Cybersecurity Grant Program and the February 2026 commitment of $10 million in application programming interface credits for cyber defense. (openai.com 1) (openai.com 2) The product launch arrived amid separate scrutiny around OpenAI leadership. Reuters reported on April 15 that Chief Executive Sam Altman asked a federal court in St. Louis to dismiss punitive-damages claims in a civil lawsuit filed by his sister, Annie Altman, while denying her allegations. (reuters.com) For now, OpenAI is not treating GPT‑5.4‑Cyber like a general ChatGPT feature. It is treating it like controlled infrastructure for vetted defenders, with access widened carefully instead of opened to the public all at once. (openai.com)