Securing Agents Isn’t the Customer’s Job, It’s the Platform’s

The article warns that as enterprises adopt AI agents, security must evolve to address data leakage, prompt-injection attacks, non-deterministic autonomous actions, and the complexity of multi-model environments; it advocates that agent and platform vendors, not just customers, must bake in strict permission boundaries, automated sensitive-data detection, grounding of AI outputs, and graduated trust models so agents can operate safely and reliably while maintaining performance.