OpenAI Expands Daybreak With GPT-5.6-Cyber Access
OpenAI expanded its Daybreak cybersecurity program on August 10, adding Blue and Red access tiers and releasing GPT-5.6-Cyber to approved defenders. The release followed OpenAI's August 7 disclosure that some Astra work was paused while stronger controls were added because preliminary evaluations could not rule out Critical cyber capability. GPT-5.6-Cyber itself was assessed at the lower High threshold.
OpenAI expanded its Daybreak cybersecurity access program on August 10, introducing two tiers and making GPT-5.6-Cyber available to approved defenders. The announcement followed the company's August 7 disclosure that it had paused some activities involving Astra, an upcoming model, until stronger security controls were in place because preliminary evaluations could not rule out the Critical cyber-capability threshold in its Preparedness Framework.
OpenAI defines Critical cyber capability as the ability to autonomously identify and develop functional zero-day exploits across many hardened real-world critical systems, or to devise and execute novel end-to-end attacks against hardened targets from a high-level goal. The company said Astra's result was preliminary and that the unreleased model was not involved in the Hugging Face incident.
Blue and Red access tiers
Daybreak Blue gives approved defenders access to frontier general-purpose models, including GPT-5.6 Sol, with system-level cyber safeguards removed for authorized defensive work. OpenAI lists vulnerability discovery, secure code review, malware analysis, incident response, and patch validation among the intended uses.
Daybreak Red provides purpose-trained cybersecurity models for authorized vulnerability research, exploit validation, and security testing. GPT-5.6-Cyber, built on GPT-5.6 Sol, is the first model announced for the tier. OpenAI says it was trained to improve specialized tasks such as zero-day discovery and exploit-chain development while reducing refusals for certain higher-risk, dual-use requests.
OpenAI's internal Advanced Cybersecurity Completion Rate evaluation found that GPT-5.6-Cyber completed 95% of advanced requests involving areas such as exploit chains, authentication bypass, and privilege escalation. The company reported 1.5% for GPT-5.6 Sol with standard safeguards and 2% through Daybreak Blue. Axios independently reported the same figures.
Those rates measure whether a model responds, not whether every answer is correct, safe, or operationally useful. OpenAI also reported that GPT-5.6-Cyber outperformed some models on exploit-development tasks but trailed GPT-5.6 Sol on one vulnerability-discovery and report-writing evaluation because its reports were sometimes shorter and less detailed.
Controlled access and operational risk
OpenAI assessed GPT-5.6-Cyber at the High cyber-capability threshold, below Critical. That separates the controlled release from Astra, whose preliminary evaluation remains under review.
Daybreak access is limited to approved individuals and organizations conducting authorized work. OpenAI says it uses identity verification, account security, monitoring, approved-use restrictions, and legal attestations. It also recommends isolated environments, explicit scope, review of elevated actions, and human oversight for higher-risk workflows.
For security teams, the practical distinction is between broad defensive assistance in Blue and more permissive exploit-validation work in Red. The response-rate gap shows why specialized access can help legitimate researchers, but it also increases the importance of authorization, sandboxing, audit logs, and independent validation before a model-generated finding changes production systems.
Key Points
- 1OpenAI added gated Blue and Red Daybreak tiers, separating general defensive assistance from specialized exploit-validation and vulnerability-research access.
- 2OpenAI reported a 95% advanced-request completion rate for GPT-5.6-Cyber, compared with 1.5% for standard GPT-5.6 Sol and 2% through Daybreak Blue.
- 3GPT-5.6-Cyber was assessed at the High cyber threshold, while Astra remains under expanded testing because OpenAI could not rule out Critical capability.
Scoring Rationale
The gated release of a specialized cyber model and the expansion of Daybreak are notable for security engineers evaluating AI-assisted vulnerability research and incident response. Astra's potential Critical capability classification also makes this a consequential public disclosure about frontier-model cyber risk controls, though access is limited to approved defenders.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
