The facts

  1. OpenAI announced GPT-6 Cyber on 5 November and said initial access is limited to a subset of verified Daybreak Red users for authorized cybersecurity work. [1][2]

  2. OpenAI classifies the model as Critical in cybersecurity under its Preparedness Framework; this is the company’s assessment, not an independent rating. [1][2]

  3. OpenAI reported reproducing 1,724 Linux vulnerabilities and 81 Firefox vulnerabilities in tests. These are company benchmarks, not confirmed real-world exploit outcomes. [1][2]

Still unconfirmed

  • Independent evaluation of the reported benchmarks and broader real-world performance is not available in the opened sources.

Sources

References support the numbered facts above. Source checks: 6 Nov 2026, 03:17 UTC.

  1. Introducing GPT-6 Cyber

    OpenAI · Primary source · 5 Nov 2026

    Opened OpenAI’s announcement and safety material; access and capability claims are attributed to the company.

  2. OpenAI launches GPT-6 Cyber with Critical cybersecurity rating

    Axios · Reporting · 5 Nov 2026

    Opened Axios’ report on the restricted rollout and company test claims.

Report a factual error