OpenAI has broadened its Daybreak program with a new purpose-built model, GPT‑5.6‑Cyber, aimed at giving vetted defenders frontier offensive-security capabilities before threat actors can weaponize AI at scale.
Daybreak Blue grants access to general-purpose frontier models, including GPT‑5.6 Sol, with safeguards tailored for authorized defensive work such as vulnerability discovery, secure code review, malware analysis, incident response, and patch validation.
Daybreak Red, by contrast, unlocks OpenAI’s purpose-trained cybersecurity models for authorized vulnerability research, exploit validation, and security testing, with GPT‑5.6‑Cyber as its flagship offering.
OpenAI Expands Daybreak With GPT-5.6-Cyber
Built on GPT‑5.6 Sol, GPT‑5.6‑Cyber is specifically trained to reduce refusals on dual-use offensive tasks including zero-day discovery and exploit-chain development that even ungated versions of GPT‑5.6 Sol decline to complete.
OpenAI’s internal Advanced Cybersecurity Completion Rate benchmark, covering exploit-chain development, authentication bypass, and privilege escalation scenarios, shows GPT‑5.6‑Cyber completing 95.0% of requests, compared to just 1.5% for safeguarded GPT‑5.6 Sol and 2.0% for Daybreak Blue access.
This also surpasses the predecessor GPT‑5.5‑Cyber’s 57.3% completion rate, directly addressing researcher complaints about persistent refusals in earlier iterations.
On ExploitGym, which tests conversion of known vulnerabilities into working code-execution exploits in isolated environments, GPT‑5.6‑Cyber outperformed both GPT‑5.6 Sol and GPT‑5.5‑Cyber.
It also led on an internal Zero-Day Discovery evaluation measuring severity calibration and technical write-up quality. Notably, it underperformed GPT‑5.6 Sol on vulnerability report writing, which OpenAI attributes to a tendency toward shorter.
On ExploitBench, which evaluates V8 exploit development against sandbox defenses, GPT‑5.6 Sol (Daybreak Blue) held an edge in token efficiency at standard turn limits, though the gap narrowed with extended turns.
Using the model against V8, Chrome’s JavaScript engine, OpenAI researchers uncovered two chainable memory-corruption vulnerabilities that could escape the heap sandbox, which they disclosed to Google and are now tracked as CVE-2026-15903.
A high-severity flaw where the optimizing compiler skipped a bounds-check safety measure during integer conversion, potentially enabling arbitrary code execution within Chrome’s sandbox.
Additional findings include five vulnerabilities in a major mobile OS (including a privilege-escalation chain from untrusted apps), three critical remote code execution issues in a widely used database, and over 400 privilege-escalation flaws in a popular OS kernel.
Under OpenAI’s Preparedness Framework, GPT‑5.6‑Cyber was rated High in cybersecurity capability, but remained below the Critical threshold. OpenAI clarified the model was not involved in the prior Hugging Face security incident.
New safeguards accompanying the release include mandatory hardware security keys for Daybreak accounts starting September 1, 2026, a push toward Codex’s auto-review mode over full-access mode, and expanded monitoring.
Access requires identity verification, legal attestations, and approved-use restrictions, with applications open at OpenAI’s Daybreak partners portal.
Detect, investigate, and respond faster with in-browser data inspection from ANY.RUN. Gain complete phishing visibility to strengthen your SOC and reduce MTTR
