OpenAI Announces New Cybersecurity Model and Two-Tier Access
GPT-5.6-Cyber offers specialized solutions for advanced security research.
1OpenAI's New Cybersecurity Strategy
On August 10, 2026, OpenAI announced the division of its Daybreak program into two access tiers and introduced GPT-5.6-Cyber, a new cybersecurity model. Daybreak Blue provides approved defenders with access to general-purpose models like GPT-5.6 Sol for daily security work, while Daybreak Red offers GPT-5.6-Cyber under stricter vetting for vulnerability research. This structure addresses the issue where system safeguards were blocking legitimate defense work.
2Capabilities of GPT-5.6-Cyber
OpenAI's new GPT-5.6-Cyber model is trained to reduce refusal rates on advanced cybersecurity tasks. It improves performance on specialized work like zero-day vulnerability discovery and exploit chain development. According to OpenAI's Advanced Cybersecurity Completion Rate metric, GPT-5.6-Cyber completes 95% of requests, compared to 1.5% for standard GPT-5.6 Sol and 2% for GPT-5.6 Sol under Daybreak Blue. The previous model, GPT-5.5-Cyber, achieved 57.3%.
3Performance Comparisons
OpenAI stated that GPT-5.6-Cyber outperformed both GPT-5.6 Sol and GPT-5.5-Cyber in ExploitGym tests. However, it fell behind GPT-5.6 Sol in Vulnerability Discovery and Report Writing evaluations, which the company attributes to shorter, less detailed reports. In ExploitBench tests, general-purpose GPT-5.6 Sol showed the best performance within the standard 300-turn limit. All data comes from internal benchmarks.
4Real-World Achievements
Using GPT-5.6-Cyber, OpenAI discovered two previously unknown vulnerabilities in Chrome's V8 JavaScript engine that could lead to memory corruption and sandbox escape. The first was reported to Google and fixed as CVE-2026-15903. The company also revealed at least five vulnerabilities in a popular mobile OS, three critical flaws in a database, and over 400 privilege escalation vulnerabilities in an OS kernel, all undergoing disclosure with partners.
5User Experiences
SpecterOps CTO Jared Atkinson, who tested the model early, described its performance as 'work compression': GPT-5.6-Cyber completed tasks in under a day that previous models couldn't resolve after weeks of intermittent effort. This feedback highlights the efficiency gains for security researchers.
6Access and Security Measures
Access to both tiers requires identity verification, account security requirements, monitoring, and usage restrictions. OpenAI will mandate hardware security keys for all individual Daybreak accounts starting September 2026. The company is also transitioning Codex users from full-access to an auto-review mode that evaluates high-permission actions before execution.
7Industry Collaborations and Past Successes
In June 2026, OpenAI launched GPT-5.5-Cyber and the Daybreak Cyber Partner Program with companies like Accenture and CrowdStrike. The open-source 'Patch the Planet' initiative collaborated with Trail of Bits. GPT-5.5-Cyber achieved 85.6% on CyberGym and 39.5% on ExploitGym, surpassing previous versions. Hundreds of issues were identified in the first five days, with dozens of patches merged.
8Balancing Cybersecurity Needs
Under OpenAI's Preparedness Framework, GPT-5.6 Sol and GPT-5.6-Cyber were rated 'High' for cybersecurity capability. This announcement followed reports that OpenAI's upcoming Astra model might cross the 'Critical' threshold. The company emphasized that GPT-5.6-Cyber wasn't involved in the Hugging Face exploitation and no models with such involvement will be released, addressing security concerns.
Original article: Unite.AI — OpenAI Siber Güvenlikte Yeni Model ve İki Katmanlı Erişim Duyurdu