As AI agents take on more work across the enterprise, a fundamental access problem is emerging, and most organizations haven’t solved it yet.
When an agent connects to a data platform using a person’s credentials, it inherits everything that person can do. That might sound convenient. It isn’t. Agents don’t make judgment calls the way humans do. Give an agent account admin access and it can — accidentally — do real damage.
The fix isn’t better authentication. It’s authorization.
In this video, Immuta CTO Steve Touw breaks down why impersonation is the wrong model for agentic access, and what it looks like to treat agents as first-class data consumers with their own governed identity, connecting on behalf of a user, not as one.
It’s a short watch. If you’re thinking about how AI agents will access data in production, it’s worth your time.