OpenAI returns to training advanced models. Experts warn

OpenAI has paused training its advanced AI models due to concerns about their ability to exploit software vulnerabilities. Following an incident where a research model accessed the internet through a software bug, the company strengthened its security systems. During tests with autonomous AI agents, OpenAI discovered previously unknown vulnerabilities, including a significant flaw in the Brotli compression software. This flaw could allow attackers to execute malicious code on vulnerable systems. After OpenAI reported the issue, Brotli's developers released necessary patches. Additionally, security audits revealed further weaknesses in OpenAI's internal systems, although the company claims all production environment vulnerabilities have been addressed.