OpenAI announced GPT‑6 Cyber on November 5, 2026, a model it describes as trained to find and validate exploitable vulnerabilities while staying within boundaries set by researchers. According to OpenAI, the model scores 48.5% on ExploitGym, a benchmark testing whether a model can exploit a specified vulnerability, up from 32.1% for the prior GPT‑5.6 Cyber. On a new offensive web penetration testing evaluation, OpenAI says it achieved 40.3% vulnerability recall versus 32.0% for GPT‑6 Astra. The company states the model avoided exploiting out-of-scope services in 99.93% of tests without a deployment prompt, and 100% with it. OpenAI reports its researchers used the model to develop 948 privilege escalation exploits in Linux and identify 81 previously unknown high-severity vulnerabilities in Firefox, and that Mozilla has released fixes for many of them. OpenAI says security teams at Cisco, Calif, and CodeRabbit are using it, and that it classifies the model as Critical in cybersecurity under its Preparedness Framework, up from High. The company says it plans to expand its Daybreak program to make security research available to organizations, independent researchers, and open-source maintainers. These figures come from OpenAI's own announcement and system card.
- GPT‑6 Cyber scores 48.5% on ExploitGym, up from 32.1% for GPT‑5.6 Cyber (OpenAI)
- Classified Critical in cybersecurity under OpenAI's Preparedness Framework, up from High
- OpenAI reports 81 previously unknown Firefox vulnerabilities found; Mozilla released fixes for many
- Model stayed within authorized scope in 99.93% of honeypot tests without a deployment prompt, per OpenAI
What it means for you
A frontier model is now capable enough at finding and exploiting software vulnerabilities that OpenAI itself labels it 'Critical' risk and is gating access to verified defenders. For most organizations this is background, not a buying decision — the practical takeaway is that the same capability that helps defenders also helps attackers, and the window where defense stays ahead is not guaranteed. If you run software that matters, the honest move is to make sure your patching and dependency hygiene are actually working, not to chase a new AI tool.
Try this
Pick one externally-facing system this week and confirm you have a current, documented process for how security patches get reviewed and applied — including how fast a critical fix reaches production. If that process lives only in one person's head, write it down.
Who should care
Whoever owns security and patching for your software and infrastructure, and teams maintaining externally-facing applications or open-source projects.
Skip this if
You don't build or maintain your own software and rely on vendors who handle security updates for you — in that case this is news to note, not act on.
Sources: OpenAI — read the original