OPENAI on Tuesday unveiled GPT-5.4-Cyber, a Defence-focused variant of GPT-5.4 designed for defensive cybersecurity use cases, as it expands access for security teams. The company said it is ramping up its Trusted Access for Cyber (TAC) programme to thousands of authenticated individual defenders and hundreds of teams securing critical software.
OpenAI also warned that AI systems are dual-use and that adversaries could invert models fine-tuned for defence to discover and exploit vulnerabilities, underscoring the need for safeguards and careful rollout. The story notes that the AI-powered application security agent from Codex Security has contributed to fixing over 3,000 critical and high vulnerabilities, and that the broader TAC expansion aims to democratise access while strengthening guardrails against jailbreaks and adversarial prompt injections.
According to OpenAI, the move follows a rival glimpse into frontier models such as Anthropic’s Mythos, which is being deployed in a controlled manner as part of Project Glasswing and is described as finding thousands of vulnerabilities across software systems.