If the Agent Cannot Say “Unknown,” It Cannot Touch the Budget.

UNKNOWN is becoming a legitimate media-buying answer.

In AAMP 2.3, IAB Tech Lab removed language-model fallbacks that could fabricate CPMs when pricing data was absent and added a pricing-provenance field. The release also introduced an optional vendor-approval gate. The significant shift is that production systems are beginning to encode when an agent must stop. Source: IAB Tech Lab’s AAMP 2.3 release

The Change Is a Refusal

AAMP 2.3 arrives after buyer and seller agents became capable of discovering inventory, negotiating, and booking deals. The release also confronts a less glamorous requirement: an agent negotiating real money must encounter missing information without manufacturing continuity. Source: IAB Tech Lab’s AAMP 2.3 release

That is the commercial event. Automation usually presents progress as a larger action surface. More channels. More transactions. More speed. Here, maturity appears as a narrower permission: no source, no price.

Analysis: capability is no longer the scarce proof. Constraint quality is. A system that assembles a plan in seconds but cannot preserve uncertainty is not faster judgment. It is faster exposure.

Provenance Replaces Plausibility

AAMP 2.3’s pricing change addresses an earlier failure mode in which an agent could supply a fabricated CPM when pricing data was missing. The new provenance field makes the source of a price traceable through the negotiation chain and allows guardrails against invented floor prices. Source: IAB Tech Lab’s AAMP 2.3 release

A plausible price answers, “Does this number look normal?” Provenance asks which agreement, rate card, seller response, or approved rule authorized it. The first is statistical resemblance. The second is commercial custody.

Analysis: once an agent can transact, plausibility becomes a dangerous substitute for authority. A number may sit comfortably inside a market range and still be unauthorized, stale, attached to the wrong inventory, or detached from negotiated terms. The control must inspect origin before reasonableness.

Optional Controls Are Not Operating Policy

The Buyer Agent now includes a vendor-approval gate connected to the IAB Diligence Platform and SafeGuard Privacy. When enabled, it filters unapproved vendors before discovery and checks again before a transaction is finalized. The official release is equally important about the limit: the gate is off by default and invisible when no key is present. Source: IAB Tech Lab’s AAMP 2.3 release

What changed is the availability of an enforceable boundary. What did not change is the organization’s responsibility to activate it, define its approved population, and decide what happens when a vendor fails the check.

The standard cannot know whether a company permits an unfamiliar seller, requires human approval above a spend threshold, or forbids a channel under certain consent conditions. It can carry controls. It cannot supply the commercial constitution.

Leverage Moves Toward the Rule Writer

AAMP’s buyer-agent architecture can ingest a brief, allocate budget, research inventory, recommend a plan, and—after approval—book deals autonomously. Its registry supports discovery of agents by purpose, category, maturity, capability, and connection method; registered companies are associated with GPP and TCF identifiers for validation. Sources: AAMP 2.0 release; IAB Tech Lab Agent Registry

Those mechanisms give governance teams consequential leverage: the ability to decide which candidates enter the consideration set, which evidence counts, and which absences force escalation.

Risk moves with that leverage. Procurement cannot treat vendor policy as a document consulted after selection. Media leaders cannot leave pricing authority implicit in a rate card folder. If the rule is not machine-readable at the decision point, it is commentary.

The Boundary Must Survive the Handoff

Agentic buying is not one decision. It is a chain: brief, discovery, comparison, negotiation, approval, booking, and later optimization. A safeguard applied only at the beginning can expire as the plan changes. A price verified during negotiation can lose its authority if inventory, timing, or deal terms move before booking.

Analysis: the control object therefore needs its own continuity. Each consequential handoff should preserve four things: the source used, the policy applied, the person or system that approved the exception, and the condition that would invalidate that approval. Logs record what occurred. A durable mandate explains why the action remained allowed.

In OrionPilot, that mandate can sit beside the campaign brief as explicit buying boundaries—approved vendors, acceptable price sources, escalation triggers, and invalidation conditions—so weekly changes do not quietly inherit authority from an earlier plan.

Write the “No” Before Connecting the Agent

The urgent decision is not whether to pilot an agent. It is whether the organization can state the refusals the pilot must enforce.

Which missing field ends the negotiation? Which price sources carry authority? Which vendors never enter discovery? Which material change reopens approval? Who may grant an exception, for how long, and where is that exception recorded?

AAMP 2.3 supplies more of the machinery required to make those answers executable, including auditable workflows, approval gates, modern authentication, and pricing provenance. Source: IAB Tech Lab’s AAMP 2.3 release

But a control that exists in software and remains disabled is only a possibility. An agent earns access to the budget at the moment “unknown” becomes a boundary it cannot talk its way around.