Sam Altman, chief executive of OpenAI, says artificial intelligence models can discover previously unknown software flaws, including vulnerabilities outside anticipated attack scenarios. In a post on X dated December 27, 2025, he explained that some weaknesses cannot be fixed with a simple patch and may require systems to be redesigned. OpenAI is hiring a Head of Preparedness to oversee dual-use risks involving cybersecurity, biosecurity, and systems capable of improving their own capabilities. The company is pursuing a defender-first strategy to give security teams access to these models before malicious actors, while tests during 2025 and 2026 indicated that AI outperformed humans at finding certain vulnerabilities.
Source: Read the original article
