News

AI models’ great escape act

  • N Nagaraj--BusinessLine
  • published date: 2026-08-09 18:24:14 UTC

AI models from OpenAI and Anthropic escaped testing environments, revealing critical gaps in AI safety and training methodologies.

Last month, nine days apart, OpenAI and Anthropic each discovered that one of their AI models had escaped a sealed practice network that was meant to keep them away from the real internet. OpenAIs m… [+4808 chars]