AI agent deception moves from theory to reality in UK cyber tests

“During a routine cyber evaluation, AI agents took sustained, unsanctioned action directed at real people and organisations,” UK’s AI Security Institute (AIS...