Summary
In this OpenAI cybersecurity keynote, company leaders and industry partners discuss what they call the “defender’s window,” the narrowing gap between frontier AI capabilities and the fast-following capabilities of open-weight models available to attackers. Speakers cite adoption figures including 26% growth in weekly Codex users for marketing tasks, a 108x increase in legal use cases, and a case study of a 650-employee company reporting 30-40% efficiency gains from running 145 AI agents alongside staff.
The centerpiece announcement is Codex Security Red, a managed penetration-testing service that lets security teams define assessment scope and rules of engagement, then run investigation agents in OpenAI-hosted sandboxes with network controls and a guardian agent reviewing all outgoing traffic. The keynote also details Daybreak Blue, a defender-focused model lineup including GPT-6 Sol and the newly added Luna, built for reviewing code, investigating alerts, and patching vulnerabilities.
Speakers frame the effort as an industry-wide coalition rather than a solo OpenAI initiative, referencing a joint letter signed by roughly 100 companies committing cybersecurity as a top corporate priority. The talk offers a detailed look at how one of the largest AI labs is positioning its most capable models for defensive security work as open-weight models close the capability gap.
📺 Source: OpenAI · Published September 28, 2026
🏷️ Format: Keynote Launch
