Summary
OpenAI unveils GPT-6 Cyber, a cybersecurity model designed to find and verify exploitable vulnerabilities. The company says it improves 51% over GPT-5.6 Cyber on ExploitGym and achieves 99.93% to 100% accuracy in avoiding unauthorized exploitation, depending on the prompt. In practice, it helped fi…
Key points
- Helps security teams understand the practical capability boundaries of AI agents in penetration testing and remediation workflows, as well as access restrictions.
- This is the first time a major model vendor has packaged high-risk offensive and defensive capabilities into a controlled product line, which will change how enterprises conduct security research and code review.
- Enterprise security leaders need to assess whether to apply for Daybreak Red access and re-examine scope governance and audit records for authorized testing.
- The company uses GPT-5.6 Cyber as the baseline and notes improvements in both ExploitGym performance and unauthorized-exploitation rates.
Editorial note
This page is Code & Chain's editorial summary of public sources. It may be prepared with AI assistance and published through an automated workflow. Refer to the original sources; this content is not investment, legal, or tax advice.