Summary
OpenAI’s “Cyber at the Frontier” presentation lays out how the company sees frontier AI models reshaping cybersecurity, and what it is doing to put them in defenders’ hands. The talk frames the moment as an acceleration of a long-running trend: the internet was never built with security in mind, and AI is now dramatically lowering the cost of finding and exploiting vulnerabilities.
The presenters describe a three-part strategy: build frontier capabilities so defenders stay ahead of what is broadly available, expand access through the Daybreak program for trusted defenders, and help organizations put these tools to work through playbooks and dedicated cyber products. They also introduce Patch the Planet, a funded effort to get vulnerabilities fixed rather than merely reported.
Concrete case studies anchor the talk. A researcher used an earlier model to find the PHP shell vulnerability, and a later model independently found a more general version of the same bug. In Firefox, a campaign with GPT-6 Cyber surfaced 81 previously unknown high-severity vulnerabilities. In Linux, an autonomous discovery, exploitation and patching pipeline validated and patched nearly a thousand local privilege escalation bugs, more than five times what GPT-5.6 Cyber found earlier. The talk also covers coordinated disclosure with companies and governments, and why chaining low-severity bugs changes patching urgency.
📺 Source: OpenAI · Published November 12, 2026
🏷️ Format: Keynote Launch
