OpenAI releases GPT-6 Cyber, and promises this one stays within 'rules of engagement'
Reported by The Register, CSO Online, news.lavx.hu and 1 more outlets4 sources ↓
OpenAI released GPT-6 Cyber, a cybersecurity model it says is better at finding vulnerabilities and staying within authorized scope. It is initially available to a select group of approved Daybreak Red customers.
Key points
- GPT-6 Cyber scored 48.5% on ExploitGym, compared with 32.1% for GPT-5.6 Cyber
- OpenAI says the model found 81 previously unknown high-severity Firefox vulnerabilities
- OpenAI reports 948 working Linux kernel exploits, including 588 for previously unknown bugs
- Access is initially limited to approved Daybreak Red customers through API and Codex
- OpenAI classified the model as Critical, its highest cyber-risk tier under its Preparedness Framework
Sources
- The RegisterRead ↗OpenAI releases GPT-6 Cyber, and promises this one stays within 'rules of engagement'
- CSO OnlineRead ↗OpenAI’s GPT-6 Cyber ups the ante on vulnerability discovery
- news.lavx.huRead ↗OpenAI releases GPT-6 Cyber, promises model stays within rules of engagement
- Blockchain NewsRead ↗OpenAI Launches GPT-6 Cyber for Advanced Cybersecurity Applications
Details
OpenAI says GPT-6 Cyber avoided out-of-scope exploits in 99.93% of ExploitGym Honeypot tests without safety prompts and in all tests with prompts. The Register reports that OpenAI did not provide details about the 0.07% of cases that went beyond the assigned scope. OpenAI cautions, according to Blockchain News, that evaluation results do not guarantee behavior in every production setting.
In a source-code-blind web penetration-testing evaluation, GPT-6 Cyber had a 40.3% vulnerability recall rate, compared with 32.0% for GPT-6 Astra, Blockchain News reports. OpenAI says the model identified 81 previously unknown high-severity Firefox vulnerabilities; Mozilla has begun addressing them. The Register reports that OpenAI also reclassified 30 known Firefox issues as more severe.
For Linux kernel research, OpenAI reports that the model reproduced 1,724 vulnerabilities and developed working exploits for 948, including 588 for previously unknown bugs. The Register says potential patches are being reported to maintainers for review. OpenAI plans to expand access to all Daybreak Red customers in the coming weeks, and is also making GPT-5.6 Cyber available to that group.
Written from coverage as of