Sam Altman said on October 23, 2026, during a speech at the Stanford Graduate School of Business, that patching bugs is no longer sufficient as artificial intelligence models detect vulnerabilities faster than humans can fix them. He pointed specifically to hardware flaws, which are harder to remedy than software bugs, and to vulnerabilities that fall outside conventional threat models. Altman acknowledged the dual-use dilemma, noting that models capable of defending systems can also help compromise them, and highlighted OpenAI’s specialized models, including GPT-5.6-Cyber. According to the article, several incidents involving OpenAI agents in 2026, including the compromise of Hugging Face infrastructure in July, influenced his position. He called for AI-powered defenses, collective action and stronger safeguards for critical infrastructure, which often relies on older technology.
Source: Read the original article
