OpenAI Expands Security Offerings with Launch of GPT‑5.4‑Cyber

OpenAI on Tuesday introduced GPT‑5.4‑Cyber, a specialized version of its newest flagship model, GPT‑5.4, designed specifically to support defensive cybersecurity operations. The announcement comes just days after rival AI firm Anthropic revealed its own frontier model, Mythos.

“The continued advancement of AI empowers defenders those tasked with protecting systems, data, and users by helping them identify and remediate vulnerabilities more quickly across the digital infrastructure that society depends on,” OpenAI said.

Alongside the model launch, OpenAI announced a significant expansion of its Trusted Access for Cyber (TAC) program. The initiative is being scaled to include thousands of verified individual security professionals and hundreds of teams responsible for safeguarding critical software ecosystems.

OpenAI acknowledged the dual‑use nature of AI, noting that while such models can significantly strengthen cyber defense, they may also be repurposed by malicious actors. A key concern is that adversaries could reverse‑engineer or misuse models optimized for defensive tasks to identify and exploit vulnerabilities in widely deployed software before patches become available, increasing risk for users.

To address this challenge, OpenAI said its strategy focuses on expanding access responsibly while reinforcing safeguards. The company emphasized a controlled, iterative rollout aimed at democratizing defensive capabilities, giving legitimate defenders an advantage, and strengthening protections against misuse, including prompt injections, jailbreak attempts, and other adversarial techniques as model capabilities grow.

“As our models become more capable, our approach is to scale cyber defense in parallel widening access for trusted defenders while continuing to enhance safety mechanisms,” OpenAI said.

The company also highlighted progress with Codex Security, its AI‑powered application security agent designed to detect, validate, and propose fixes for software vulnerabilities. According to OpenAI, the tool has contributed to the remediation of more than 3,000 critical and high‑severity vulnerabilities.

GPT‑5.4‑Cyber’s limited rollout follows Anthropic’s recent preview of Mythos, which is being deployed under a tightly controlled framework known as Project Glasswing. Anthropic has stated that Mythos identified “thousands” of vulnerabilities across operating systems, web browsers, and other widely used software platforms.

OpenAI emphasized the importance of continuous security integration during software development. “A resilient ecosystem is one that consistently discovers, validates, and addresses security flaws as code is written,” the company said. “By embedding advanced coding models and agent‑based capabilities directly into developer workflows, we can provide immediate, actionable feedback moving security from periodic audits and static bug lists to continuous, measurable risk reduction.”