Why can’t we just keep rogue AIs off the internet?

Researchers debate isolating AI agents from the internet to prevent harmful behavior, but strict air gapping limits realistic testing and slows development. Balancing safety and practical evaluation remains a challenge.