Aembit CEO Proposes New Security Framework for AI Agents

David Goldschlag, CEO and co-founder of Aembit, proposes a shift in how enterprises manage identity for AI agents, arguing that current systems rely too heavily on simple identification. He warns that treating an AI agent's identity as a permission slip creates significant security risks, especially when agents possess over-scoped credentials for final resources.

Goldschlag advocates for a design centered on granular authorization and access control. In this framework, credentials would be scoped to specific hops within a workflow—such as the transition from an agent to a gateway and then to a server—rather than granting broad entitlements across the entire chain.

To further protect enterprise data and infrastructure, he recommends the use of short-lived credentials. He also emphasizes the necessity of maintaining strict audit trails that can clearly distinguish between the acting AI agent and the human user on whose behalf the agent is operating.