[KongchangAI]
ModelGPT-6Cyber

GPT-6Cyber

A model introduced by OpenAI specifically for the cybersecurity domain, used to discover and validate exploitable vulnerabilities. It is restricted to operating within boundaries set by researchers and is released in tiers according to a security framework.

Timeline (last 90 days)

Nov 8

GPT-6Cyber is used to find and verify exploitable vulnerabilities, and is limited to working within the scope set by researchers.

Unverified50%
Nov 8

Under the OpenAI safety framework, GPT-6Cyber's cybersecurity capability is rated "Zero-Tier," versus "Advanced" for the previous generation.

Unverified50%
Nov 8

Official tests show that GPT-6Cyber scored 48.5% on the remote vulnerability exploitation benchmark, about 16 percentage points higher than the previous generation.

Unverified50%
Nov 8

Official tests show that GPT-6Cyber found 81 new vulnerabilities in open-source software that the Firefox browser developer rated as high-severity, and provided reproducible tests and patch suggestions.

Unverified50%

All Facts (4)

Source Articles