EDB says AI agent controls should be enforced at the database

In short: EDB says companies should control what AI agents can do by enforcing rules directly in the database where data lives.

What's going on

More companies are using AI agents, meaning software that can plan tasks and take actions across different tools without a person approving every step. EDB says this creates a basic safety question, if an agent tries to do something it is not allowed to do, what actually stops it.

The article argues that written policies and “guardrails” added on top of an agent are not enough when the agent is acting quickly and unpredictably. It uses a simple example: “Never open the car door” sounds safe, until the car is on fire and someone needs to get out. The point is that rules often depend on context, meaning what is happening right now.

EDB’s proposal is to make governance “executable,” meaning the system can automatically allow or block actions at the moment they happen. In practice, EDB says the best place to do this is the data layer, which usually means the database (the system that stores and serves a company’s information). If the database refuses a request, the agent cannot “talk its way” into getting sensitive data, like an employee trying to enter a locked room without a badge.

What to watch

As agents get more autonomy, more companies may treat agents like separate users with their own identities and stated purpose, and then log what they did for later review. Buyers will likely ask vendors whether these controls work consistently across on-site systems and cloud systems, and whether the audit logs can clearly show which agent accessed which data and why.

Source: Venturebeat