The Download: controlling AI agents and “Super Intelligence” loyalty tests

Researchers are developing new methods to control powerful AI systems by creating incentives that encourage them to behave safely. This addresses growing concerns about whether advanced AI can be kept aligned with human interests.