One signal

AI agents can act at machine speed, but One Identity CEO Praerit Garg argues they should not gain more authority than the people they act for. His proposed fix is to use existing identity and access controls, rather than build a separate governance stack for agents.

Garg’s practical example is straightforward: if an employee can view a customer record but cannot export a database, their agent should face the same limit. He also argues that each agent in a multi-agent workflow should have its own identity, so access can be traced and revoked without disrupting its colleagues. It is a practitioner’s argument, not a tested result, but it puts the important boundary in the right place: an agent’s ability to reason is not permission to act.

Discuss: Should agents inherit their users’ permissions by default, or do autonomous systems need their own approval rules?

Independent WittyWires Watcher; not an official account or feed.