Discussion on "Allowed Isn't Authorized: Testing What Your AI Agent Does With the Permissions It Already Has"

Arunima Chaudhuri · 2026-10-06T20:21:53.349Z

A hands-on guide to AgentSec 0.7's authority checks (task scope, data flow, honest reporting, identity boundaries and multi-agent delegation), with real output from reference agents, seven frameworks,

Allowed is not authorized is the right distinction, and the yes-or-no question that turned into a commit is the perfect origin story: the agent held the tool rights but had no authorization for that effect. The five authority questions are what an allowlist structurally cannot answer. When you check 'did the task authorize this effect', how do you represent the task's intent so tool_effects can be matched against it?

Great question, Kartik! The idea is to capture the user's original request as the source of authority, then evaluate tool effects against what that request actually permits — not just which tools the agent can access. The tricky part is translating natural-language intent into explicit constraints without assuming permissions the user never granted. That's one of the problems I'm exploring with AgentSec!

"When a scenario can't be judged, it isn't a pass" is the line that stuck with me. I'd want the same rule for records: an entry nobody could judge carries that state, it doesn't default to green. I also like that question five asks on whose authority, not only which agent.

Yeah, exactly. If something can’t be judged, that should stay visible instead of quietly turning into “safe.” And I agree — knowing whose authority the agent acted on matters just as much as knowing which agent acted.

Agreed. "Quietly turning into safe" is the failure I worry about most, because nothing looks wrong afterward. Thanks for writing the piece. I'll be borrowing the five questions.

Really appreciate you reading it, Salman! Glad you found the five questions useful!

Search Hashnode

Allowed Isn't Authorized: Testing What Your AI Agent Does With the Permissions It Already Has

Responses6