Everything in this track assumes the buyer agent is talking to a seller that is actually allowed to sell the inventory. That assumption is exactly where fraud lives. Before a buy, an agent has to confirm authorization, the same check you ran by hand in Track C, now between agents.
A property owner publishes adagents.json to declare which sales agents may represent its properties. A buyer agent reads it, and uses list_authorized_properties to confirm a given sales agent is cleared for the property it is offering. Only then does it deal. This is the agentic counterpart of ads.txt plus the sellers.json reciprocity check, one layer up.
| Trust question | Header bidding | AdCP |
|---|---|---|
| Who may sell this property? | ads.txt authorized sellers | adagents.json authorized agents |
| Confirm the seller is cleared | sellers.json reciprocity | list_authorized_properties |
| Unverified party in the path | Unauthorized reseller | Unauthorized sales agent |
Watch out
A sales agent offering a property it is not listed for in adagents.json is the agentic version of an unauthorized reseller in ads.txt. It is the same red flag the Bidcliq audit raises today, and the reason authorization is checked before any money moves.
Quick check
A sales agent offers inventory for example.com, but example.com’s adagents.json does not list it. Should the buyer agent transact, and which Track C concept does this mirror?
That closes the loop on AdCP. You can describe the protocol, walk a buy through the Media Buy module, get a creative approved, activate a signal, and verify authorization. Track I covers the IAB Tech Lab’s parallel approach to the same problems.