OpenAIAnalysisOpenAI classifies GPT-6 Cyber as critical under Preparedness Framework
What happened.OpenAI classified its specialized GPT-6 Cyber model as Critical under its internal Preparedness Framework after the system demonstrated autonomous vulnerability discovery and exploit generation capabilities.
Why it matters.Automated exploit generation shifts software vulnerability economics from slow manual research to continuous machine-scale probing. Security teams must accelerate automated patch verification pipelines because offensive agents can weaponize exposed software flaws within hours.
The technical specifications:
- Exploit benchmark: The model scored 48.5 percent on ExploitGym, improving by sixteen points over its predecessor.
- Flaw validation: Automated research pipelines reproduced 1,724 Linux kernel defects and generated 948 working privilege escalation exploits.
- Zero-day discovery: The system identified 81 previously unknown high-severity vulnerabilities in Mozilla Firefox that engineers subsequently patched.
- Scope containment: Post-training guardrails achieved a 99.93 percent out-of-scope target avoidance rate during automated penetration tests.
- Access control: Production access remains restricted to approved security organizations participating in the Daybreak Red verification program.
STORYLINE PROGRESSION & CONTEXT
This report represents a verified evolution of prior newsroom intelligence: