OpenAI released GPT-6 Cyber November 5, The Register reports. Its bug-finding claims aren’t proof that customer-facing software is patched.
OpenAI says researchers identified 81 previously unknown high-severity Firefox vulnerabilities and developed exploits for 948 Linux kernel vulnerabilities, including 588 previously unknown bugs. Reuters reports 304 critical Linux findings disclosed with proposed patches; TechCrunch puts the Linux hunt’s cost at $1.2 million.
According to The Register, initial API and Codex access is restricted to selected approved Daybreak Red customers, with expansion across that program planned over the coming weeks. OpenAI rates it Critical under its Preparedness Framework, the highest cyber-risk tier, covering functional zero-day exploit development in real-world critical systems and end-to-end cyberattacks. CSO reports OpenAI recommends separate Daybreak Blue for most defenders.
The Register reports an ExploitGym score of 48.5%, versus GPT-5.6 Cyber’s 32.1%. Separately, CSO reports OpenAI’s scope-compliance evaluation: unauthorized boundary crossings in 0.07% of cases versus 5.51% for GPT-5.6 Cyber. OpenAI says there were none with deployment safety prompts.
CSO reports Mozilla has begun releasing fixes; Linux findings accompany candidate patches for maintainer review. OpenAI identified 10 apparently novel techniques in 166 exploits that could give local users root privileges, saying external kernel exploitation experts hadn’t seen them before.
TechCrunch separately reports Charge Robotics co-founder and chief technology officer Max Justicz used GPT 5.6 Sol through Daybreak Blue to find an Ubuntu server-takeover flaw. He said it took two hours. The Debian Project issued a patch; TechCrunch reported no known exploitation evidence.
OpenAI also worked with Cloudflare on PHP web application firewall rules before disclosure, CSO reports. “We wanted to get ahead of that public patch and release.” said Clint Gibler, OpenAI cybersecurity research lead.
But OpenAI provided no measurements of faster enterprise patch deployment or reduced exposure. Noma Security vice president of security strategy Chris Hughes said OpenAI’s Patch the Planet dashboard showed “a less than 25% remediation rate thus far.” That’s a dashboard figure, not GPT-6 Cyber’s measured patching benefit.
For small B2B teams, owners should ask website developers, hosting providers, and commerce vendors which software they manage, whether findings affect it, and how they verify released fixes reached production. Ask browser managers about Firefox updates; ask providers about affected Linux components and patch status. This doesn’t assume every business runs vulnerable software.
Before adding assurances beside inquiry forms, booking pages, or checkouts, get evidence. A security page can identify verified update responsibilities and a prospect contact, not recast pending patches as protection.
“AI-powered” is not an answer to a prospect asking what protects their information.
Sources
- https://openai.com/index/introducing-gpt-6-cyber/
- https://www.investing.com/news/stock-market-news/openai-launches-gpt6-cyber-with-critical-cybersecurity-rating-5005229
- https://techcrunch.com/2026/11/05/openai-released-a-new-cybersecurity-model-is-that-enough/
- https://www.theregister.com/security/2026/11/05/openai-releases-gpt-6-cyber-and-promises-this-one-stays-within-rules-of-engagement/5307855
- https://www.csoonline.com/article/4244911/openais-gpt-6-cyber-ups-the-ante-on-vulnerability-discovery.html

