OpenAI Introduces GPT-5.6-Cyber to Empower Security Defenders With AI
OpenAI has recently launched a robust cybersecurity-focused AI model, GPT-5.6-Cyber. This launch is accompanied by a significant expansion of its Daybreak Cyber Partner program.
The unveiling comes in the wake of the company’s decision to halt internal work on its forthcoming Astra model. This decision was prompted by safety evaluations that found the model perilously close to reaching a “Critical” cybersecurity threshold. This means it could independently identify and exploit zero-day vulnerabilities in hardened real-world systems without human intervention.
Built on the GPT-5.6-Sol architecture, GPT-5.6-Cyber is designed for advanced, authorized cybersecurity work. It boasts a lower refusal rate for risky dual-use tasks. According to OpenAI, the model has already been used to identify over 400 vulnerabilities in a popular OS kernel, three critical database flaws, and at least five vulnerabilities in a major mobile operating system. The model has been assessed as reaching the High — but not Critical — cybersecurity capability threshold under OpenAI’s Preparedness Framework.
The expanded Daybreak program now operates in two tiers:
- Daybreak Blue: Provides access to GPT-5.6-Sol without standard cyber guardrails for defensive tasks like vulnerability discovery and malware analysis.
- Daybreak Red: Grants vetted red teams access to GPT-5.6-Cyber for exploit validation and advanced vulnerability research.
This strategic move positions OpenAI in direct competition with Anthropic’s Project Glasswing initiative.
Source: SecurityWeek – OpenAI Unveils New Cybersecurity Model GPT-5.6-Cyber