threatdeck
Live 14:45 UTC ·

Model Watch Medium For CISO, SOC / IR, AppSec, AI Eng, Red Team AI-assisted

OpenAI releases GPT-6 Cyber, a restricted cybersecurity model for approved defenders

OpenAI introduced GPT-6 Cyber, which scores 48.5% on ExploitGym versus 32.1% for GPT-5.6 Cyber. OpenAI reports it stayed within authorized scope in 99.93% of alignment tests without a deployment prompt, and 100% with one. Access is initially limited to approved Daybreak Red customers through the API and Codex.

Why it matters

Dual-use models that can develop working exploits raise the risk of misuse by malicious actors, and organizations that use them need controls on scope and access. Defenders gain faster vulnerability discovery, but the source gives no independent verification of OpenAI's benchmark figures.

Do this

  1. Review whether your organization has any Daybreak Red or GPT-6 Cyber access, and confirm that API keys and credentials for these services are managed under your standard secrets policy.
  2. If you use Daybreak Red, enforce the default IP allowlisting and monitor credentials for compromise, as OpenAI describes.
  3. Run any autonomous security agent in a sandbox with action review and human oversight for high-risk workflows, as OpenAI recommends.
  4. Track Mozilla Firefox and Linux kernel security advisories so fixes for the disclosed vulnerabilities can be applied to your systems.

Signals

  • NoCISA KEV
  • 22/100Priority
  • 60%Confidence

Evidence

“On ExploitGym, which tests whether models can exploit a specified vulnerability, GPT‑6 Cyber scores 48.5%, up from 32.1% for GPT‑5.6 Cyber.”

OpenAI News

“GPT‑6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers.”

OpenAI News

“We require IP allowlisting for Daybreak Red API access by default, restricting requests to approved networks with limited exceptions.”

OpenAI News

Sources

Affected / related

GPT-6 Cyber, Daybreak Red, Daybreak Blue, GPT-5.6 Cyber, GPT-6 Astra, OpenAI, Mozilla

Published 2026-11-05 09:00 UTC · updated 2026-11-10 00:24 UTC · AI-assisted summary, checked against quoted sources · How we score

Following

Follow the vendors, products, AI tools, techniques and actors you care about. Matching stories, CVEs and exploited bugs are marked, a Following tab appears in the feed, and returning visits show what changed. Saved only in this browser.

Presets

Get updates

A personal RSS feed of new stories on everything you follow. Paste it into any feed reader, Slack or Teams RSS app.