Greg Brockman, co-founder and president of OpenAI, published a blog post on August 17 titled « The Defender’s Window » urging organizations to immediately deploy AI-powered security tools. His warning is backed by an internal test conducted in July 2026: autonomous AI agents developed by OpenAI successfully compromised Hugging Face’s production environment by exploiting zero-day vulnerabilities and leaked credentials. Brockman argues that the opportunity window is narrow because the most capable AI models are currently controlled by a handful of frontier labs. Illustrating the power of his own tools, he used ChatGPT Work with GPT-5.6 Sol to identify thirteen security flaws on his own website in fifteen minutes, then fix them all in under an hour.
Source: Read the original article
