OpenAI Launches GPT-5.6-Cyber Through Daybreak for Advanced Security Research

OpenAI has expanded its Daybreak cybersecurity programme with two access tiers and introduced GPT-5.6-Cyber, a model trained for tasks including zero-day discovery, exploit development, and security testing.

The company said Daybreak Blue will give approved defenders access to GPT-5.6 Sol with safeguards for authorised security work, while Daybreak Red will provide access to purpose-trained cybersecurity models. GPT-5.6-Cyber will be available through Daybreak Red.

“Threat actors will increasingly use AI to conduct cyberattacks at unprecedented speed and scale, including in fully autonomous ways,” OpenAI said. The company added that defenders have a “narrowing window” to prepare for these capabilities.

GPT-5.6-Cyber is built on GPT-5.6 Sol and trained for specialised cybersecurity tasks, including finding zero-day vulnerabilities and developing exploit chains. OpenAI said the model also reduces refusals for certain higher-risk, dual-use cybersecurity tasks.

In an internal evaluation called Advanced Cybersecurity Completion Rate, OpenAI tested how often models responded to requests involving exploit-chain development, authentication bypass, privilege escalation, and other advanced cybersecurity scenarios. GPT-5.6-Cyber completed 95% of these requests, compared with 1.5% for GPT-5.6 Sol and 2% for GPT-5.6 Sol with Daybreak Blue access. GPT-5.5-Cyber recorded a 57.3% completion rate.

OpenAI said Daybreak Blue is intended for vulnerability discovery, secure code review, malware analysis, incident response, patch validation, and security assessments. Daybreak Red is intended for authorised vulnerability research, exploit validation, and security testing.

The company also said GPT-5.6-Cyber performed better than GPT-5.6 Sol and GPT-5.5-Cyber on ExploitGym2, an evaluation that tests whether AI agents can turn known vulnerabilities into working exploits in controlled environments.

OpenAI said it used GPT-5.6-Cyber to investigate V8, the JavaScript engine used by Chrome, and found two previously unknown vulnerabilities that could be chained to corrupt memory and escape the V8 heap sandbox.

The researchers validated the findings and reported them to Google through coordinated vulnerability disclosure. Google subsequently fixed the vulnerability, which OpenAI said was assigned CVE-2026-15903.

OpenAI also said GPT-5.6-Cyber helped identify at least five vulnerabilities in a mobile operating system, three critical vulnerabilities in a database, and more than 400 vulnerabilities that could lead to privilege escalation in an operating system kernel.

The company said GPT-5.6-Cyber was assessed as reaching the High threshold under its Preparedness Framework, but not the Critical threshold for cybersecurity capability.

Daybreak Blue and Daybreak Red are available to approved individuals and organisations conducting authorised cybersecurity work. OpenAI said access will be controlled through identity verification, account security, monitoring, approved-use restrictions, and legal attestations.

OpenAI is also requiring individual Daybreak accounts to use hardware security keys from September 1. The company said it is working on additional monitoring measures and is encouraging customers using Codex to use auto-review mode for actions requiring elevated permissions.

OpenAI recommends Daybreak Blue as the starting point for most defenders, while organisations involved in advanced vulnerability research, exploit development, or red teaming can request Daybreak Red access.