Who can get Daybreak Red access?

Daybreak Red is open to approved individuals and organizations whose authorized work includes vulnerability research, exploit development, or red teaming. OpenAI controls entry through identity verification, account security, monitoring, approved-use restrictions, and legal attestations. Teams apply through the Daybreak partners page. OpenAI recommends Daybreak Blue as the starting point for most defenders instead.

How is Daybreak Blue different from normal GPT-5.6 Sol?

Daybreak Blue gives approved defenders GPT-5.6 Sol without the system-level guardrails that screen cybersecurity requests in production. Those guardrails block real defensive work, so Blue removes them for vulnerability discovery, secure code review, malware analysis, incident response, and patch validation. GPT-5.6 Sol still refuses highly dual-use prompts such as pentesting production systems, even under Blue.

What real vulnerabilities has GPT-5.6-Cyber found?

GPT-5.6-Cyber found two previously unknown flaws in V8, the JavaScript engine in Chrome, that could be chained to corrupt memory and escape the V8 heap sandbox. Google fixed the issue as CVE-2026-15903. OpenAI also reports three critical database bugs, at least five mobile OS flaws, and over 400 privilege-escalation issues in an OS kernel.

Was GPT-5.6-Cyber involved in the Hugging Face breach?

No. OpenAI states directly in the Daybreak announcement that GPT-5.6-Cyber was not involved in exploiting Hugging Face, and that no other models are planned for an upcoming release in connection with it. The company points back to its earlier Hugging Face incident updates for the full account of what happened.

Is GPT-5.6-Cyber better than GPT-5.6 Sol at every security task?

No. GPT-5.6-Cyber beats GPT-5.6 Sol on ExploitGym 2 and on zero-day severity calibration, but scores worse on OpenAI's internal Vulnerability Discovery and Report Writing evaluation — OpenAI believes the model sometimes writes shorter, less detailed reports. On ExploitBench with a 300-turn limit, GPT-5.6 Sol under Daybreak Blue performs best.

OpenAI · 2026-08-10 · major

GPT-5.6-Cyber — OpenAI splits Daybreak into Blue and Red tiers

GPT-5.6-Cyber is OpenAI's new security model, built on GPT-5.6 Sol and gated behind a Daybreak Red tier. It answers 95.0% of advanced cyber requests, against 1.5% for GPT-5.6 Sol with normal safeguards.

OpenAI Daybreak announcement card for GPT-5.6-Cyber

OpenAI ships a security-specific model and puts it behind a vetted tier so approved defenders stop hitting refusals.

Key specs

Advanced cyber completion rate	95.0%
Chrome v8 cve found	CVE-2026-15903

Quick facts

Maker	OpenAI
Base model	GPT-5.6 Sol
Access tiers	Daybreak Blue and Daybreak Red
Where GPT-5.6-Cyber lives	Daybreak Red only
Availability	Approved individuals and organizations doing authorized work
Login requirement	Hardware security keys from September 1, 2026
Preparedness rating	High for cyber capability, below Critical

Benchmarks

Advanced Cybersecurity Completion Rate

GPT-5.6-Cyber (Daybreak Red)		95%
GPT-5.5-Cyber		57.3%
GPT-5.6 Sol (Daybreak Blue)		2%
GPT-5.6 Sol (safeguards on)		1.5%

source ↗

What is it?

GPT-5.6-Cyber is a new cybersecurity model trained on top of GPT-5.6 Sol for finding zero-day vulnerabilities and building exploit chains. OpenAI released it alongside a two-tier restructure of its Daybreak program. Daybreak Blue hands approved defenders frontier general-purpose models with the cyber guardrails removed; Daybreak Red gates the purpose-trained models behind tighter vetting.

How does it work?

The split is about refusals, not raw capability. In production, system-level safeguards screen cybersecurity requests, and they also block legitimate defensive work — so Daybreak Blue turns them off. Some dual-use prompts still get refused by the base model, so OpenAI trained GPT-5.6-Cyber to answer them and to perform better on exploit-chain development, authentication bypass, and privilege escalation.

Why does it matter?

Security teams have spent a year fighting model refusals on work they are paid and authorized to do. Daybreak now gives them a documented path: Blue for day-to-day defense, Red for research that needs the sharper model. The Chrome CVE and the kernel findings show the capability is real, which is also why access is gated and hardware keys become mandatory.

Who is it for?

security researchers, incident responders, red teams

Frequently asked questions

Who can get Daybreak Red access?: Daybreak Red is open to approved individuals and organizations whose authorized work includes vulnerability research, exploit development, or red teaming. OpenAI controls entry through identity verification, account security, monitoring, approved-use restrictions, and legal attestations. Teams apply through the Daybreak partners page. OpenAI recommends Daybreak Blue as the starting point for most defenders instead.
How is Daybreak Blue different from normal GPT-5.6 Sol?: Daybreak Blue gives approved defenders GPT-5.6 Sol without the system-level guardrails that screen cybersecurity requests in production. Those guardrails block real defensive work, so Blue removes them for vulnerability discovery, secure code review, malware analysis, incident response, and patch validation. GPT-5.6 Sol still refuses highly dual-use prompts such as pentesting production systems, even under Blue.
What real vulnerabilities has GPT-5.6-Cyber found?: GPT-5.6-Cyber found two previously unknown flaws in V8, the JavaScript engine in Chrome, that could be chained to corrupt memory and escape the V8 heap sandbox. Google fixed the issue as CVE-2026-15903. OpenAI also reports three critical database bugs, at least five mobile OS flaws, and over 400 privilege-escalation issues in an OS kernel.
Was GPT-5.6-Cyber involved in the Hugging Face breach?: No. OpenAI states directly in the Daybreak announcement that GPT-5.6-Cyber was not involved in exploiting Hugging Face, and that no other models are planned for an upcoming release in connection with it. The company points back to its earlier Hugging Face incident updates for the full account of what happened.
Is GPT-5.6-Cyber better than GPT-5.6 Sol at every security task?: No. GPT-5.6-Cyber beats GPT-5.6 Sol on ExploitGym 2 and on zero-day severity calibration, but scores worse on OpenAI's internal Vulnerability Discovery and Report Writing evaluation — OpenAI believes the model sometimes writes shorter, less detailed reports. On ExploitBench with a 300-turn limit, GPT-5.6 Sol under Daybreak Blue performs best.

Try it

Apply to join the program at openai.com/daybreak/partners