IAB AI Transparency and Disclosure Framework (2026)
Tags: Frameworks
TL;DR
- Two‑layer model: clear consumer‑facing disclosure when material, plus internal/machine‑readable provenance for all AI‑involved assets.
- Decisions are anchored by a materiality test focused on authenticity, identity, representation, and decisions that influence consumers.
Why it matters for HK marketers: It gives a practical, defensible way to decide when and how to disclose AI involvement without over‑labeling everything.
Two-Layer Disclosure Model
- Layer 1 — Consumer‑facing notice: Used where lack of disclosure could cause deception or confusion; tailored to the medium (on‑screen text, audible notice, opening bot message, etc.).
- Layer 2 — Provenance record: Maintain machine‑readable or internal records (e.g., C2PA, platform labels, asset IDs, generation records) indicating how the asset was created and who approved the decision.
Materiality Test and Decision Flow
- Ask whether AI makes a person appear to say/do something that did not occur; creates a photorealistic person/event/product result/testimonial; simulates a human interaction; or materially changes a representation that could influence decisions.
- Distinguish assistance vs. synthetic representation: routine production assistance is generally low disclosure risk; fully generated or materially manipulated realistic content needs closer review.
- Decision steps:
1) Was AI involved in creating/materially altering the asset/interaction?
2) Could that involvement change reasonable consumer understanding?
3) Would absence of disclosure risk deception/harm? If yes, disclose before release and embed provenance.
4) Is content lawful/accurate/authorised? If not, do not publish—labels do not cure illegality or falsity.
Execution and Recordkeeping
- Make disclosures clear and proximate in plain language.
- Record wording, placement, rationale, approver, and any platform label applied; verify post‑launch that labels persist after reformatting.
- Preserve provenance metadata and avoid stripping it; use persistent watermarking for higher‑risk synthetic media.
High-Risk Scenarios Needing Escalation
- Deepfakes, digital twins, cloned voices, synthetic humans in primary roles.
- Fabricated events/locations, AI‑generated testimonials, manipulated news‑like content.
- Political/public‑interest communications, vulnerable audiences, customer‑facing agents easily mistaken for people, or inconsistent platform label requirements.
So what for marketers
Adopt the IAB decision flow in your pre‑launch QA, require asset‑level provenance retention, and apply prominent disclosures wherever materiality is met.
Sources:- (June 14) Responsible AI in Digital Marketing Playbook - Executive Summary & Checklists.pdf
← Back to Knowledge Base