Should an always-on agent have its own identity and accounts?

Once it holds a standing role rather than acting as an extension of one person, yes: OpenAI gives its specialist dots their own identity and credentials, and Google’s coworker agents get a dedicated identity down to their own @agents.company.com email address, specifically so a later audit trail says the agent did something rather than a person.

That’s a deliberate split, not a default. OpenAI’s everyday dot works as an extension of you, under access you’ve connected and permissions you manage; a specialist dot, set up by your company for a defined role, gets its own identity, credentials, and access instead. Google’s coworker agent keeps its own persistent storage too, and sees only what your team shares with it, rather than inheriting whatever the person who set it up could already reach.

OWASP’s 2026 Top 10 for agentic applications names the alternative as its own risk category, ASI03, Identity and Privilege Abuse: without a governed identity, an agent sits in what the framework calls an attribution gap, where real least privilege isn’t enforceable. A dedicated identity is also what makes the record an agent leaves say which agent did something, not just who was logged in when it ran.