Quick Answer
GPT-6 Cyber is OpenAI’s first cybersecurity model classified as Critical under its Preparedness Framework, and it is initially restricted to approved Daybreak Red customers. OpenAI says the model improves authorized vulnerability discovery, exploit development, web pentesting, and fix validation. Organizations seeking access need identity and security controls, while most users should treat its benchmark results as self-reported rather than independent proof.
Key Takeaways
- OpenAI announced GPT-6 Cyber on November 5, 2026.
- GPT-6 Cyber is classified as Critical, compared with High for GPT-5.6 Cyber.
- OpenAI reports a 48.5% ExploitGym score, up from 32.1% for GPT-5.6 Cyber.
- Approved Daybreak Red customers need SSO, MFA, identity verification, and API IP allowlisting.
- GPT-6 Cyber supports a 1.05-million-token context window and costs $20 per million input tokens.
What did OpenAI launch with GPT-6 Cyber?
OpenAI launched GPT-6 Cyber as a specialized model for authorized cybersecurity work, including vulnerability discovery, exploit development, web pentesting, and validating security fixes. The November 5, 2026 release is initially available only to approved Daybreak Red customers, rather than through broad public access. OpenAI’s launch announcement describes the model as a tool for security professionals working within authorized environments.
GPT-6 Cyber matters because the model is designed to support the full security-testing workflow instead of only summarizing code or explaining known vulnerabilities. A security team can use a model in that role to investigate a suspected weakness, develop a controlled proof of concept, and test whether a proposed patch resolves the issue. The authorization requirement remains important because exploit-development capabilities can create real risk when used against systems without permission.
GPT-6 Cyber is separate from general-purpose GPT-6 products. Readers comparing OpenAI’s broader lineup should distinguish the restricted cybersecurity release from the company’s broader GPT-6 strategy, which covers a different product and access context. The practical conclusion is that consumers and most individual developers should not expect ordinary chatbot access to GPT-6 Cyber at launch.
Why is GPT-6 Cyber classified as Critical?
GPT-6 Cyber is classified as Critical under OpenAI’s Preparedness Framework, while GPT-5.6 Cyber received a High classification. OpenAI assigns that classification based on the model’s cybersecurity capability profile and the safeguards it says are necessary before deployment. OpenAI’s GPT-6 Cyber system card sets out the Critical designation and the access controls attached to the model.
The Critical classification does not mean that GPT-6 Cyber is available for unrestricted offensive use. The designation means OpenAI considers the model’s cybersecurity capability serious enough to require stronger deployment safeguards than the prior High-rated model. That distinction matters because a model that can help authorized teams validate defenses can also lower the effort needed for harmful activity if access controls fail.
OpenAI’s classification is a company assessment, not an independent government certification or a guarantee that every use will be safe. The most sensible interpretation is that organizations should evaluate the model as a controlled security capability with substantial governance requirements. Security teams should also maintain human review, written authorization, and ordinary incident-response procedures when using AI-assisted testing.
How much better did GPT-6 Cyber perform in testing?
GPT-6 Cyber scored 48.5% on ExploitGym, compared with 32.1% for GPT-5.6 Cyber, according to OpenAI. OpenAI describes the difference as a 51% relative improvement. The result indicates stronger performance on the company’s exploit-related evaluation, but the figure is self-reported and not a substitute for independent testing.
| Measure | GPT-6 Cyber | GPT-5.6 Cyber | What the comparison shows |
|---|---|---|---|
| ExploitGym score | 48.5% | 32.1% | OpenAI reports a 51% relative improvement for GPT-6 Cyber. |
| Preparedness classification | Critical | High | OpenAI applies stronger safeguards to the newer model. |
| Access approach | Approved Daybreak Red customers | Not specified in the released comparison | GPT-6 Cyber begins with restricted organizational access. |
OpenAI also says GPT-6 Cyber reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 of them. The company says 588 of those vulnerabilities were previously unknown, and that it disclosed 304 critical findings to maintainers with proposed patches. Reuters reported the figures provided by OpenAI, which show the company is presenting the model as capable of finding and validating serious software flaws.
GPT-6 Cyber also found 81 previously unknown Firefox vulnerabilities that Mozilla classified as high severity, according to OpenAI, along with 30 known issues that OpenAI says were more severe than previously reported. Those figures are consequential because browser vulnerabilities can affect widely used software, but they do not establish how consistently the model will perform across every codebase, configuration, or security environment. Organizations should validate findings through experienced security engineers before treating an AI-generated result as confirmed.
What access controls does GPT-6 Cyber require?
GPT-6 Cyber access requires identity verification, organization-level security requirements, single sign-on, multifactor authentication, and API IP allowlisting. OpenAI says the controls apply to Daybreak Red customers approved to use the model, and it plans to expand access to additional approved customers in the coming weeks. The requirements make the initial release more restrictive than a standard developer API rollout.
Single sign-on, commonly called SSO, connects access to an organization’s identity system. Multifactor authentication, or MFA, requires more than a password before a user can sign in. API IP allowlisting restricts requests to approved network addresses. Together, those controls help organizations identify users, reduce credential-based misuse, and limit where API requests can originate.
GPT-6 Cyber’s access model is particularly relevant after recent security incidents involving exposed credentials and account sessions. Organizations using AI systems with sensitive internal context should treat access governance as part of the security work itself, especially when responding to breaches such as the Chainalysis password and token exposure. The practical response is to limit access to named users, review API logs, and remove access promptly when a security role changes.
What are GPT-6 Cyber’s context window and API prices?
GPT-6 Cyber supports a 1.05-million-token context window and a maximum output of 128,000 tokens, according to OpenAI’s API documentation. OpenAI lists pricing at $20 per million input tokens, $2 per million cached input tokens, and $100 per million output tokens. OpenAI’s API model documentation provides the published specifications and pricing.
The 1.05-million-token context window allows GPT-6 Cyber to work with large amounts of supplied material in a single request, such as source code, security documentation, test outputs, and vulnerability reports. That capability can be useful when an investigation spans several repositories or a large application environment. A large context window does not guarantee that the model will interpret every relationship correctly, so human reviewers still need to confirm the model’s conclusions.
The output price matters because exploit analysis, patch validation, and detailed security reports can require lengthy responses. Organizations should estimate both input and output usage before treating the model as a routine automated scanner. Teams that need application-risk tooling should also compare how AI assistance fits into established vulnerability-management processes, including approaches discussed in application security platform comparisons.
What does GPT-6 Cyber mean for software maintainers?
GPT-6 Cyber means software maintainers may face a faster and more capable source of vulnerability reports from authorized security teams. OpenAI says the model identified critical Linux kernel findings with proposed patches and found previously unknown high-severity Firefox vulnerabilities. The reported work suggests that AI-assisted research can help surface flaws that might otherwise take longer to identify.
Software maintainers still need to reproduce, prioritize, and fix every reported issue through normal security processes. A model-generated proof of concept can be useful evidence, but maintainers need to assess exploitability, affected versions, patch quality, and potential regressions. The appropriate response is careful validation rather than automatic acceptance of an AI-generated report.
GPT-6 Cyber also increases the importance of prompt security updates when maintainers confirm a vulnerability. End users cannot evaluate exploit chains themselves, but they can reduce exposure by installing patches from trusted vendors. That same principle applies to browser updates, including the urgency behind recent critical Chrome security fixes.
Who should use GPT-6 Cyber, and who should not?
GPT-6 Cyber is intended for approved organizations conducting authorized security research, penetration testing, vulnerability validation, and defensive engineering. The model’s access requirements indicate that OpenAI is targeting security teams that can meet organizational controls, rather than individual users seeking a general coding assistant. The model may be useful where teams already have authorization boundaries, responsible-disclosure procedures, and staff qualified to verify results.
GPT-6 Cyber is not an appropriate tool for testing systems without explicit permission. Unauthorized scanning, exploit development, or attempts to access another party’s network can harm victims and create legal consequences, even if an AI model helped produce the technical material. The safe action is to test only systems your organization owns or has written authorization to assess.
Organizations should stop and involve their security leadership or legal team before using GPT-6 Cyber on production systems, customer environments, or third-party infrastructure. Security teams should also maintain backups and rollback plans before testing proposed fixes in production-like environments. AI assistance can speed research, but it does not remove the need for accountable human approval.
FAQ
What is GPT-6 Cyber?
GPT-6 Cyber is OpenAI’s specialized model for authorized vulnerability discovery, exploit development, web pentesting, and security-fix validation. OpenAI announced the restricted model on November 5, 2026.
Is GPT-6 Cyber available to the public?
GPT-6 Cyber is not broadly available to the public at launch. OpenAI initially limits access to approved Daybreak Red customers that meet identity and organization-level security requirements.
Why did OpenAI rate GPT-6 Cyber as Critical?
OpenAI rated GPT-6 Cyber as Critical because the company determined its cybersecurity capabilities require stronger safeguards than a High-rated model. The classification is OpenAI’s preparedness assessment, not an independent certification.
How much does GPT-6 Cyber cost?
GPT-6 Cyber costs $20 per million input tokens, $2 per million cached input tokens, and $100 per million output tokens under OpenAI’s published API pricing. Organizations should estimate output use carefully because detailed security analysis can generate long responses.
Can GPT-6 Cyber replace a human penetration tester?
GPT-6 Cyber cannot replace a human penetration tester because security findings still require validation, authorization checks, risk assessment, and accountable remediation decisions. Qualified security professionals should review every AI-generated exploit, vulnerability report, and proposed patch.