OpenAI launches GPT-6 Cyber, rated Critical on cybersecurity by its own framework
A model trained to find and validate exploitable flaws while staying in scope. OpenAI says researchers used it to build 948 Linux privilege escalation exploits and find 81 unknown high-severity Firefox flaws. It rates the model Critical under its Preparedness Framework, up from High for GPT-5.6 Cyber. No independent test results were found.
Why it matters: It is a strong offensive tool in restricted hands, and all of its scores currently come from the company selling it.
- Price
- Not disclosed
- Context
- Not disclosed
- Availability
- Through OpenAI's Daybreak Red program for authorized security work
| Benchmark | Score | Type | Measured or published by |
|---|---|---|---|
| ExploitGym | 48.5% (GPT-5.6 Cyber: 32.1%) | Self-reported | OpenAI |
| Web penetration testing, vulnerability recall | 40.3% (GPT-6 Astra: 32.0%) | Self-reported | OpenAI |
| ExploitGym Honeypot, no safety prompt | 99.93% stayed in scope (GPT-5.6 Cyber: 94.49%) | Self-reported | OpenAI |
Sources: openai.com