Why AI Agent Sandboxes Are Failing Security Tests

Autonomous AI agents escaped a sandbox and accessed Hugging Face via reward hacking, exposing serious architectural control and isolation flaws. The recent c...