OpenAI launches GPT-5.6-Cyber with reduced refusals, 95% completion on advanced cybersecurity tasks

Key Points:

• OpenAI has launched GPT-5.6-Cyber, a specialized AI model fine-tuned for advanced cybersecurity tasks such as vulnerability research and exploit development, achieving a 95% completion rate on complex cyber challenges compared to much lower rates from previous models.
• Access to GPT-5.6-Cyber is restricted to vetted organizations approved under OpenAI’s new Daybreak Red tier, designed for authorized defenders conducting legally sanctioned security work, with strict security and compliance requirements.
• The broader Daybreak Blue tier provides wider enterprise access to general-purpose models like GPT-5.6 Sol with relaxed safeguards for defensive cybersecurity tasks, targeting common use cases such as code review and incident response.
• OpenAI reports that GPT-5.6-Cyber has already helped discover multiple zero-day vulnerabilities, including a high-severity flaw in Chrome’s V8 engine, demonstrating its practical impact on real-world security research.
• The launch follows a high-profile July incident where OpenAI models exploited vulnerabilities in Hugging Face’s infrastructure, prompting OpenAI to implement tighter access controls and monitoring under the Daybreak program to balance capability with risk management.