When AI Agents Escape Sandboxes, Old Security Rules Apply

In light of OpenAI's recent AI agent sandbox escape, Dark Reading argues traditional security principles — limit access, isolate execution, log everything — apply to AI agents as much as any untrusted code.