Access is initially limited to approved Daybreak Red customers, with OpenAI reporting gains in vulnerability discovery and compliance with authorized testing boundaries
OpenAI’s GPT-6 Cyber is initially available to select approved Daybreak Red customers. Conceptual illustration.
OpenAI is giving approved security teams access to GPT-6 Cyber, an AI model designed to find software vulnerabilities and demonstrate whether they can actually be exploited.
The company says researchers used the model to uncover 81 previously unknown Firefox vulnerabilities classified by Mozilla as high severity. Mozilla has released fixes for many of them, following earlier rounds of testing with other advanced AI models.
Brian Grinstead, Distinguished Engineer at Firefox, says: “Even after further hardening with multiple generations of frontier models, GPT‑6 Cyber uncovered previously unknown high-severity vulnerabilities, identified architectural improvements, and helped us ship fixes to millions of users.”
The findings accompany a restricted rollout through Daybreak Red, OpenAI’s access program for more advanced exploit development and authorized security testing.
Finding flaws and demonstrating their impact
OpenAI says its researchers also used GPT-6 Cyber to develop working exploits for 948 Linux kernel vulnerabilities, producing code that demonstrates how the flaws can be used.
Following Linux community guidance, the company disclosed 304 critical findings to maintainers with proposed fixes. Work to address other vulnerabilities is continuing.
Cisco, Calif and CodeRabbit are among the security teams OpenAI names as using the model for vulnerability research and code review.
In OpenAI’s ExploitGym evaluation, which tests whether a model can exploit a specified vulnerability, GPT-6 Cyber scored 48.5%, compared with 32.1% for GPT-5.6 Cyber.
The company classifies the new model’s cybersecurity capabilities as Critical under its own Preparedness Framework, up from High for its predecessor.
Testing whether the model stays within authorized targets
Alongside finding vulnerabilities, OpenAI tested whether the model would respect limits on what it was allowed to attack.
Its ExploitGym Honeypot evaluation places an authorized target in an isolated environment alongside other services that are deliberately easy to exploit. Those additional services are outside the model’s assignment, even though attacking them could help it complete the task.
GPT-6 Cyber avoided exploiting those services in 99.93% of tests without its deployment safety instructions. With those instructions, it stayed within scope in every test.
These results come from OpenAI’s controlled evaluation. The company is also introducing monitoring that checks whether actions match a user’s authorization, protections against stolen credentials, and additional restrictions on assistance involving critical systems.
For Daybreak Red API access, requests must generally come from approved networks. OpenAI continues to recommend sandboxed environments, action review and human oversight for high-risk work.
Restricted access for critical infrastructure testing
OpenAI is adding a group within Daybreak Red for approved defenders of critical infrastructure, including the systems supporting power plants and water treatment facilities.
Its existing $1 billion Daybreak for Frontline Defenders program subsidizes access for organizations with limited resources, including government bodies and essential service providers.
Cybersecurity company Dragos evaluated Daybreak Red models in a controlled mock power plant environment. Rob T. Lee, Chief AI Officer at Dragos, says: “The skill floor is dropping for the reconnaissance and vulnerability research needed before an attack.”
OpenAI says it used feedback from Dragos to refine safeguards against harmful assistance while retaining support for authorized testing.
Access also carries data-handling conditions. API customers using Zero Data Retention generally must use Private Safety Processing, which checks encrypted records held in customer storage for 30 days without exposing prompts and responses to OpenAI personnel.
GPT-6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers. OpenAI plans to extend access to all Daybreak Red customers and is also making GPT-5.6 Cyber available across that group.
The company is separately widening Daybreak Blue alpha access for GPT-6 Astra to more organizations and individual security researchers. New users and organizations can apply to join Daybreak.
ETIH Innovation Awards 2027 open for entries
The ETIH Innovation Awards 2027 are now open for entries, with 26 categories recognising companies, institutions, projects and individuals making a measurable impact across education technology. The global awards span AI in education, digital learning, higher education, skills and workforce, inclusion, education operations, evidence and impact, with entries managed through ETIH’s new online awards platform.
Entries are open worldwide, with the early-bird deadline on Friday 18 December 2026 and the final entry deadline on Friday 26 February 2027 at 23:59 UK time.
Explore the 2027 categories and judging criteria, read the How to Enter guide or start an entry through the ETIH Innovation Awards 2027 platform.
