Agentic AI systems could threaten security and data privacy, unless organizations test each model and component