OpenAI releases GPT-5.4-Cyber, specifically designed to lower the response refusal threshold for security personnel, requiring identity verification before use

ME News Report, April 15 (UTC+8), according to Beating Monitoring, OpenAI released GPT-5.4-Cyber, a model variant based on GPT-5.4 specifically fine-tuned for cybersecurity scenarios. The main difference from the standard version is that it proactively lowers the rejection boundary for security-related requests, allowing security personnel to perform operations that the general model would normally refuse, such as binary reverse engineering—analyzing compiled software for malicious code or security vulnerabilities without source code. This model is not open to all users but is deployed in layers through OpenAI’s “Trusted Access for Cyber” (TAC) program. GPT-5.4-Cyber is only available to top-tier users, including vetted security vendors, corporate security teams, and security researchers. OpenAI also stated that users with access to more permissive models may face additional restrictions, especially in low-visibility scenarios like zero data retention (ZDR), with stricter limits when accessed via third-party platforms. The TAC program itself is expanding, currently reaching thousands of individual defenders and hundreds of security teams. Individual users can verify their identity at chatgpt.com/cyber to gain basic tier access with reduced rejection, while enterprises can apply through an OpenAI account representative. OpenAI claims this layered strategy will guide future testing and deployment of more powerful models: standard models with general safety measures will be widely released, while specialized, more permissive models trained for security scenarios will always be deployed in a restricted manner. (Source: BlockBeats)