The debate comes at a fraught moment for the administration. AI agents are increasingly able to browse the web, operate software and write code with limited human supervision, and have escaped human control in a string of high-profile incidents. Used by a hostile actor, they could also identify vulnerabilities or automate cyberattacks at a speed that outpaces existing safeguards. Zuckerberg’s involvement is not the first time the administration’s AI policymaking process has faced last-minute dra...