OPENAI has launched a new cybersecurity-focused AI model, GPT-5.6-Cyber, alongside an expansion of its Daybreak Cyber Partner program. This model is tailored for advanced cybersecurity tasks, boasting a significantly lower refusal rate for sensitive dual-use operations compared to its predecessor, GPT-5.6-Sol. OpenAI claims a 95% success rate in executing complex tasks like exploit chains and vulnerability assessments.
To mitigate risks of misuse, GPT-5.6-Cyber will only be available to trusted partners within the Daybreak program, which now features two access tiers: Daybreak Blue and Daybreak Red. The launch follows concerns regarding the potential threats posed by autonomous AI systems, especially as the new Astra model from OpenAI potentially nears critical cybersecurity risks.