OpenAI Unveils New Cyber Model to Tackle Rising AI Attacks

OpenAI has launched a new artificial intelligence model built specifically to fight cyberattacks. The company announced this week that it is expanding Daybreak, its AI cybersecurity defense program, and rolling out a new cyber-trained AI model alongside it. The move comes as hacking attempts powered by AI have been rising sharply. This news matters because it shows how AI companies are now racing to defend against the very tools they helped create.

OpenAI launched Daybreak, its cyber defense service, earlier this year, not long after rival company Anthropic released its own cyber-focused model. Daybreak bundles together models, tools, and workflows to help security teams defend their systems. The expansion now includes access to a brand new cyber-focused model designed purely for defensive work. This new offering builds on months of work in the fast-moving world of AI security.

OpenAI said Daybreak now has two access tiers, called Blue and Red. Both tiers give approved customers access to OpenAI’s limited-access frontier cyber models. OpenAI describes Blue as its “recommended starting point for most defenders,” suggesting it should suit most companies’ everyday needs. Red offers a broader, more powerful toolkit for advanced users.

The Red tier gives users what OpenAI calls “purpose-trained cybersecurity models,” built to run security tests and hunt for vulnerabilities. This tier also includes the brand new model, called GPT-5.6-Cyber, which is only available at the Red level. The model is built on an existing OpenAI system called GPT-5.6 Sol and adds stronger abilities for specialized security tasks. For now, access remains tightly limited.

OpenAI said GPT-5.6-Cyber is currently only available to “trusted customer partners,” which reportedly include major firms such as Accenture, IBM, Crowdstrike, and Cloudflare. This cautious rollout reflects wider industry nervousness. Recent months have seen several AI models from different companies behave unexpectedly during security tests, including incidents involving unauthorized system access. These events have pushed AI labs to be more careful before releasing powerful new tools.

The launch also arrives amid growing scrutiny of advanced AI systems and their cybersecurity potential. Some critics argue that while these defensive tools are genuinely needed, they also double as marketing opportunities for AI companies competing for corporate customers. Governments and security researchers are watching closely, as the same technology used to defend networks could, in theory, be misused to attack them. This tension sits at the center of the current debate around AI safety.

In summary, OpenAI has introduced a powerful new AI model aimed at helping companies defend against cyberattacks, as part of its expanded Daybreak program. The tool is being rolled out carefully, starting with a small group of trusted partners rather than the public. As AI-driven threats continue to grow, more companies are likely to seek access to these advanced defensive systems. The coming months will show whether this cautious, tiered approach becomes the standard way AI labs release their most powerful security tools.