OpenAI Launches GPT-5.6-Cyber as AI Security Threats Grow

OpenAI is expanding its Daybreak cyber defence service and introducing GPT-5.6-Cyber, a specialised model designed for advanced cybersecurity work, as AI systems become increasingly capable of finding and exploiting software vulnerabilities.

The expansion comes amid growing attention to AI-led security incidents, including cases involving systems accessing unintended networks, exploiting vulnerable websites and using social engineering techniques. AI developers are responding by expanding tools intended to put similar capabilities in the hands of cybersecurity defenders.

OpenAI expands Daybreak with Blue and Red tiers.

OpenAI launched Daybreak earlier this year as a cyberdefence service combining access to AI models, tools and workflows for security professionals. The company said Monday that it is expanding the program into two tiers, called Daybreak Blue and Daybreak Red.

Both tiers provide approved customers with access to OpenAI’s limited-access frontier cyber models. The company has previously placed significant safeguards around its most capable cybersecurity systems because of the potential for the same capabilities to be used offensively.

Daybreak Blue is positioned as the starting point for most security teams. It includes capabilities for incident response, malware analysis and patch validation, among other defensive cybersecurity tasks.

OpenAI describes Blue as its “recommended starting point for most defenders,” indicating that the tier is designed to cover common enterprise security needs without providing the broader capabilities available through Red.

GPT-5.6-Cyber arrives with Daybreak Red

Daybreak Red provides access to a more advanced set of cybersecurity capabilities. OpenAI says the tier includes purpose-trained cybersecurity models intended for authorised security testing and vulnerability research.

The centrepiece of Red is GPT-5.6-Cyber, a new specialised model available only through that tier. OpenAI says GPT-5.6-Cyber is based on GPT-5.6 Sol and has enhanced capabilities for specific cybersecurity tasks.

Access is initially restricted to trusted customer partners. According to the source reporting, companies participating include Accenture, IBM, CrowdStrike and Cloudflare, among others.