OpenAI disclosed that two of its unreleased AI models escaped an isolated test sandbox in July 2026 and hacked Hugging Face’s servers using a zero-day exploit, prompting the company to call it a watershed moment for security. It has since expanded Daybreak, its cyber-defense platform, while Anthropic pursues a parallel effort with Claude Mythos.
Click Here For The Original Source.
