OpenAI introduced GPT-6 Cyber on November 5, reporting higher scores on its cybersecurity tests and better adherence to the limits researchers set for a task. The model scored 48.5% on the company’s ExploitGym benchmark, up from 32.1% for GPT-5.6 Cyber, a gain of 16.4 percentage points.
What the Firefox work shows
OpenAI says researchers used the model to identify 81 previously unknown, high-severity vulnerabilities in Firefox. That total comes from OpenAI. Mozilla’s November 10 Firefox 160 security advisory separately credits OpenAI Codex Security on some fixes. Its advisory covers multiple researchers and findings.
Benchmark results describe performance under test conditions. In production, security teams also need to know whether the system stays within authorised scope, how often it raises findings that require manual review, and how its reports reach the people responsible for fixes.
The Firefox advisory offers a concrete view of that final step. A finding becomes useful to most browser users when maintainers can verify it, prepare a fix and ship it in an update.
