AI Agent Security: Whose Responsibility Is It?

AI agent security hinges on a shared responsibility model, with both vendors and customers needing robust guardrails as agentic AI deployments become indispensable for productivity. The piece notes that data is not stored directly in an AI agent but in enterprise repositories to which agents are granted access, so securing that data is the customer’s responsibility even when vendors provide the infrastructure.

It highlights that incidents such as Salesforce’s ForcedLeak, a vulnerability chain in Agentforce, illustrate how improper access controls or prompt injections can lead to data exfiltration, underscoring the risk of leakage if security reviews are skipped. Experts quoted in the article argue that vendors should secure the infrastructure while customers must secure data and users, and that tools like secrets scanning or data loss prevention cannot fully solve the underlying data access problem.

The author urges organisations to prioritise security, understand data flows, implement guardrails, and recognise the ongoing challenge of clearly delineating responsibility in agentic AI environments.