OpenAI's Secret Weapon: A New AI Model Hunts Hackers Before They Strike

OpenAI has quietly released a powerful new cybersecurity tool to a select group of customers. The model, named GPT-5.4-Cyber, is designed to find software vulnerabilities before criminals can exploit them. This move comes as competition heats up in the AI security field. Rival company Anthropic recently launched its own specialized model, called Mythos, for the same purpose. Both tools aim to automate the hunt for software bugs, a critical task usually done by human experts. The limited release suggests OpenAI is testing GPT-5.4-Cyber's real-world performance. The company has not announced when, or if, it will be available to all customers. If successful, such AI models could significantly strengthen digital defenses. They would allow companies to find and fix security flaws faster than ever, potentially stopping major cyber attacks before they begin.