OpenAI's GPT-5.4-Cyber: Exclusive Access for Cybersecurity Experts

Global - Ekhbary News Agency

OpenAI has recently introduced GPT-5.4-Cyber, a specialized iteration of its large language model tailored exclusively for cybersecurity tasks. Unlike its general-purpose counterparts, this version will not be made available to the wider public via ChatGPT. Instead, OpenAI is initiating a 'Trusted Access for Cyber' program, granting limited access to verified cybersecurity professionals. The objective of this initiative is to rigorously test the model, identify potential vulnerabilities, and enhance its defensive capabilities against sophisticated cyber threats. Feedback from these experts will be crucial for improving the model's resilience against jailbreaks and other adversarial attacks. This move underscores the escalating competition in the AI and cybersecurity landscape, particularly following recent announcements from rivals like Anthropic's Project Glasswing. GPT-5.4-Cyber features lower guardrails for security-related tasks, enabling experts to comprehensively assess its potential and the risks of its weaponization by malicious actors. This strategic decision highlights OpenAI's commitment to security and the responsible deployment of advanced AI technologies.

Ekhbary News Agency