SAN FRANCISCO: OpenAI launches GPT-5.6-Cyber, a model trained for advanced cybersecurity work, with access restricted to vetted defenders.
The company announces the model on August 10 alongside an expansion of Daybreak, its cyber defense program. Daybreak now operates in two tiers. Daybreak Blue provides GPT-5.6 Sol with system-level cyber guardrails removed, supporting vulnerability discovery, malware analysis, incident response, and patch validation. Daybreak Red provides GPT-5.6-Cyber for exploit validation and advanced vulnerability research.
OpenAI says the model addresses high refusal rates that have limited security teams. In company testing, standard GPT-5.6 Sol answered 1.5% of advanced security requests involving exploit chains, authentication bypass, and privilege escalation. GPT-5.6-Cyber answered 95% of the same requests.
The company says “defenders have a narrowing window to prepare” before attackers deploy offensive AI at scale. OpenAI reports using the model internally in vulnerability research, including work that uncovered previously unknown flaws in Chrome’s V8 engine.
Sixteen partners gain access to embed the models in security products and managed services. The group includes CrowdStrike, Palo Alto Networks, Cisco, IBM, and Accenture. All individual Daybreak accounts must adopt hardware security keys by September 1.
GPT-5.6-Cyber reached the “High” capability threshold under OpenAI’s Preparedness Framework. Axios reports the company recently delayed a separate model, Astra, after it crossed critical hacking thresholds during safety testing.
The launch follows the UK AI Security Institute’s August 4 report documenting unsanctioned actions by frontier AI agents during testing. Anthropic offers similar defender access through its Glasswing program, and startup Corma raised $60 million this week for defense-focused models.
Previous
