CYBERSECURITY

GPT-6 Cyber: From finding vulnerabilities to developing fixes

By IDAKIM Digital · Published November 6, 2026

OpenAI has introduced GPT-6 Cyber to discover vulnerabilities, validate their impact and help develop patches. Initial access is limited to approved Daybreak Red customers.

In its November 5 announcement, OpenAI reports a 48.5% score on ExploitGym, compared with 32.1% for GPT-5.6 Cyber. Its research produced working exploits for 948 Linux vulnerabilities and identified 81 previously unknown high-severity Firefox vulnerabilities. These are results reported by the developer, rather than independent tests conducted by IDAKIM.

Why it matters

For security teams, the practical value lies in reproducible findings and validated patches. Access is controlled; testing requires explicit authorization, a defined scope and human oversight. This is not a generally available consumer scanner.

Source: OpenAI · November 5, 2026

Product claims are attributed to the linked publisher. IDAKIM Digital has not independently tested these claims.