OpenAI Launches GPT-5.4-Cyber for Defensive Security and Expands Professional Access

OpenAI has launched GPT-5.4-Cyber, a specialized version of its flagship GPT-5.4 model fine-tuned for defensive cybersecurity tasks, complete with expanded access for thousands of verified professionals. This move directly responds to rival Anthropic's recent restriction of its more powerful Mythos model, intensifying competition in AI-driven cyber defense.

The new model features lower refusal boundaries for legitimate cybersecurity work, allowing it to handle sensitive tasks without triggering standard safeguards that might block dual-use activities. It introduces advanced capabilities like binary reverse engineering, enabling analysts to scrutinize compiled software for vulnerabilities, malware, or weaknesses without source code access. According to OpenAI's announcement, this positions defenders to identify and fix flaws faster than attackers can exploit them.

Alongside the model release, OpenAI is scaling its Trusted Access for Cyber (TAC) program, originally introduced in February 2026, to include thousands of individual defenders and hundreds of teams safeguarding critical software. Access operates through a tiered system: individuals can verify at chatgpt.com/cyber, while enterprises contact OpenAI representatives. Approved users gain entry to cyber-permissive models for security education, defensive programming, and vulnerability research, with GPT-5.4-Cyber rolling out iteratively to vetted vendors, organizations, and researchers.

This expansion builds on OpenAI's prior tools, such as the Codex Security agent, which has already helped fix over 3,000 critical and high-severity vulnerabilities. The company emphasizes a balanced approach: democratizing access while strengthening safeguards against misuse, jailbreaks, and prompt injections as models grow more capable. OpenAI frames the initiative as preparation for even more advanced models expected later this year.

The timing underscores a heated rivalry with Anthropic, which last week limited Mythos to select users—a decision OpenAI portrays as less accessible compared to its "democratized" TAC framework. As reported by TechRadar, GPT-5.4-Cyber serves as OpenAI's direct counter, promising broader reach for pros hunting next-level threats in programs and apps.

Cybersecurity experts and organizations stand to benefit most, gaining tools to bolster defenses for critical infrastructure amid rising AI-powered attacks. What happens next includes ongoing rollouts, higher-tier authentications for TAC participants, and OpenAI's promised upgrades, all aimed at giving defenders a proactive edge in an evolving threat landscape. This development matters broadly, as it could accelerate vulnerability remediation across industries while navigating the risks of powerful AI in sensitive domains.