OpenAI Launches GPT-5.6 Cyber to Automate Zero-Day Discovery
The new Daybreak program grants vetted security firms access to a model capable of constructing complex exploit chains.
OpenAI has released GPT-5.6 Cyber, a specialized model designed to automate advanced cybersecurity tasks, including zero-day discovery and exploit chain construction. The release marks a strategic shift toward segmenting AI capabilities to provide high-utility tools to security professionals while maintaining strict public safeguards.
The model is deployed through the "Daybreak" program, which is split into two distinct tiers. "Daybreak Red" provides gated access to GPT-5.6 Cyber for approved researchers, while "Daybreak Blue" gives vetted defenders access to GPT-5.6 Sol with reduced system-level cyber guardrails. In internal tests, GPT-5.6 Cyber answered 95% of requests involving privilege escalation, authentication bypass, and exploit chains—a massive leap over GPT-5.6 Sol, which answered only 1.5% to 2% of similar requests.
Controlled Access and Validation
To mitigate the risk of misuse, OpenAI has implemented rigorous onboarding for Daybreak users. Participants must verify their identities, sign legal attestations, and accept continuous monitoring. Starting September 1, 2026, the program will also require the use of hardware security keys. Early access has already been granted to a consortium of industry leaders, including SpecterOps, SentinelOne, and Palo Alto Networks, with further partnerships extending to IBM, Accenture, Cisco, and CrowdStrike.
Proven Offensive Capabilities
The model's efficacy has already been demonstrated through the discovery of significant vulnerabilities. GPT-5.6 Cyber identified two previously undisclosed flaws in Chrome's V8 JavaScript engine, which directly led to the fix for CVE-2026-15903. Additionally, the model uncovered three critical vulnerabilities in a database and more than 400 privilege-escalation flaws within an operating system kernel. Jared Atkinson, CTO of SpecterOps, noted that GPT-5.6 Cyber completed work in less than a day that previous models had failed to resolve after several weeks.
Industry Implications
This release represents a significant escalation in AI-driven offensive security. By automating the discovery of zero-day vulnerabilities, OpenAI is effectively lowering the technical barrier for constructing sophisticated attacks. However, by gating these tools, the company aims to empower defenders to find and patch flaws before they can be exploited by malicious actors. Under OpenAI's Preparedness Framework, GPT-5.6 Cyber is currently rated as having "High" cybersecurity capability, though it remains below the "Critical" threshold that would trigger more severe restrictions.
What to Watch
Industry observers will be monitoring whether the "Daybreak" model of gated access can prevent the leakage of these capabilities into the wild. While the current framework keeps the model below the critical risk threshold, the rapid pace of discovery suggests that the line between "High" and "Critical" capability may blur as the model is further refined by its early partners.