One agent, two users: whose account performs the action?
Shared reading from the original source. Read the piece, then bring your experience or a different take.
Victor Moreira
LangChain · 2026-09-09Imagine two colleagues asking the same agent to open a ticket. The task text can be identical while the identity behind the action needs to differ.
Victor Moreira’s September 9 Connections post separates two choices that are easy to mix up: who owns the credential and what kind of credential it is. The owner can be the deployed agent or the caller. The credential can be a static secret or an OAuth grant.
That gives a clearer design conversation than “we support OAuth.” A shared account may suit a common search service. A tool that changes each person’s records needs a deliberate decision about whose access it uses.
A small verification exercise: have two test users request access to different synthetic records. Inspect the acting account, revoke one user’s grant and repeat the request. Check the denied path too.
Where do you want a shared service identity, and where must the action remain attributable to the person who requested it?
Have a result, correction, or a different constraint?
Start a related discussionRelated topics
- What security questions should every AI agent vendor answer clearly?Security
- What security questions should every AI agent vendor answer clearly?Security
- How do you evaluate prompt-injection risk in customer-facing agents?Security
- Which AI tools are safest for companies with strict data residency needs?Security
- What permissions model should an internal AI agent use?Security