GPT 5.4 Cyber: Enhancing Cybersecurity with OpenAI’s New Model

GPT 5.4 Cyber marks a significant advancement in AI-driven cybersecurity initiatives. OpenAI announced on March 14 the expansion of its Trusted Access for Cyber (TAC) programme, introducing a specialised variant of GPT-5.4 designed to strengthen AI-powered cyber defences. This new model aims to empower security professionals in implementing advanced protective measures.

This announcement follows a week after Anthropic revealed Claude Mythos as part of Project Glasswing, illustrating the escalating competition among tech giants to create robust tools for identifying vulnerabilities, addressing emerging threats, and enhancing digital security.

What Makes GPT 5.4 Cyber Unique?

The standout feature of GPT 5.4 Cyber lies in its binary reverse engineering capabilities. This allows experts to scrutinise software, identify potential malware threats, uncover security weaknesses, and evaluate the robustness of applications without needing access to the source code. Here is everything pertinent about GPT 5.4 Cyber.

Understanding GPT 5.4 Cyber

The GPT‑5.4‑Cyber model has been fine-tuned specifically for AI-backed cyber defences, coming with fewer restrictions on its capabilities. This version of GPT‑5.4 is programmed to respond to legitimate requests relating to cybersecurity.

This model seeks to facilitate advanced defensive tasks like pinpointing vulnerabilities in large-scale codebases, reasoning through intricate software frameworks, analysing malware behaviour, and conducting binary reverse engineering even without source code access.

Control and Accessibility

Due to the model’s expansive capabilities, OpenAI is exercising caution regarding its deployment. Access is currently limited to trusted security firms, organisations, and researchers. According to the company, access to permissive and cyber-capable models may come with restrictions, particularly concerning confidential uses like Zero-Data Retention (ZDR).

GPT 5.4 Cyber vs Anthropic’s Claude Mythos

Both the GPT 5.4 Cyber and Claude Mythos focus on enhancing cybersecurity; however, they differ in their development. GPT 5.4 is a specialised adaptation of the original GPT 5.4 model, while Anthropic’s Claude Mythos is entirely a new model created from scratch.

Currently, Mythos is available in a preview phase to a select group of organisations involved in Project Glasswing for cyber defence testing and is not yet accessible to the broader public. In contrast, GPT 5.4 Cyber operates under a controlled access framework initiated with its TAC programme, which was unveiled in February 2026. This programme allows the verification of identities for organisations and researchers seeking to access exclusive models.