Preventing Rogue AI Agents Starts With Measuring What They Actually Do

Security researchers Bruce Schneier and Barath Raghavan argue stopping AI agents from going rogue requires measuring whether they do what we actually mean, not just what we literally said.