OpenAI Releases GPT-5.4-Cyber Model for Limited Cybersecurity Testing

OpenAI unveiled GPT-5.4-Cyber with limited access through its Trusted Access for Cyber program, positioning it as a response to Anthropic's Claude Mythos. The model is designed for defensive cybersecurity work including reverse engineering, vulnerability analysis, and malware detection, with fewer constraints on probing methods than standard GPT-5.4.