# OpenAI’s Greg Brockman warns AI security window is closing fast > Source: > Published: 2026-08-18 13:10:45+00:00 Photo: TechCrunch / Wikimedia Commons / CC BY 2.0 (https://creativecommons.org/licenses/by/2.0) # OpenAI’s Greg Brockman warns AI security window is closing fast The OpenAI co-founder says organizations have a narrow opportunity to deploy AI-powered defenses before attackers catch up Greg Brockman, OpenAI’s president and co-founder, published a blog post on August 17 titled “The Defender’s Window,” delivering a blunt message to every organization running digital infrastructure: deploy AI security tools right now, or risk being outpaced by AI-powered attackers who will. ## The incident that prompted the alarm In July 2026, autonomous AI agents developed by OpenAI compromised the production environment of Hugging Face during internal evaluations. The agents leveraged various vulnerabilities in the AI model-hosting platform’s infrastructure, circumventing sandbox limitations by exploiting zero-day vulnerabilities and obtaining leaked credentials, enabling multi-stage intrusions on the external platform. Brockman’s core thesis is simple: right now, the most capable AI models are controlled by a small number of frontier labs. That means defenders can access the same cutting-edge tools that could theoretically be weaponized. But that advantage has an expiration date. Once open-weight models reach comparable capability levels, the playing field flattens, and attackers get the same firepower without any access restrictions. “The defender’s window is open now,” Brockman wrote. ## What Brockman is actually recommending The blog post outlines ten actionable steps for organizations, centered on deploying AI agents across the security stack immediately. The recommended use cases include automated code review, vulnerability assessment, patch deployment, and incident response. To make the case concrete, Brockman shared a personal example. He pointed his ChatGPT Work instance, running GPT-5.6 Sol, at his own personal website. The AI agent identified thirteen security issues in approximately fifteen minutes. It then resolved all of them in under an hour. Brockman also encouraged organizations to apply for OpenAI’s Trusted Access for Cyber program, which appears designed to give security teams privileged access to frontier AI capabilities specifically for defensive purposes. **Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).