Why did you target me?
Trace the audience data to its supplier, applicable permission and actual use. Personal data is in scope when it supports an advertising decision.
Developing work · IAE Governance Lab · IEEE SA knowledge partnership
AI advertising should show its work.
From intention to evidence: connect what an AI ad agent was authorized to do, what people were told and what actually happened.
I’m Jay Mandel. I work with Sean Meier on Technical Conformance, Layer 4 of the developing AI Advertising Trust Stack. This page shares our contribution, examples, progress, and the questions still open.
Apply the thinking to your business
Build customer trust through reliable delivery.
My independent Trust Stack consulting service helps you trace a customer problem across promises, people, workflows, technology, and ownership. Start with a focused diagnostic, then scope the work to put the fixes into practice.
Explore the consulting service ↗The consulting service does not confer IAE or IEEE certification or endorsement. The workstream and its proposed technical capabilities remain under development.
Summit I to Layer 4
An ad agent can hit its target and exceed its authority.
The IAE Summit I report of 22 April 2026 describes agents that can change their methods while pursuing a goal. Its concerns include consumer choice, meaningful human intervention and responsibility that remains with people and companies.
An AI ad agent is software that creates ads, selects audiences or spends budget. Layer 4’s proposed mandate is to preserve evidence so actual behavior can be checked against authorized intent, permissions and disclosures.
What changed?
Compare changes in content or agent behavior with the approved task and limits. Retain human check-ins and interventions.
Who could stop it?
Record the responsible owner and the authority behind approval. Presence alone does not establish control.
Layer 4 supplies evidence. Other layers define the requirements and response. These examples do not make every use of personal data subject to the same legal consent requirement. See the report and workstream inputs in the source register ↗.
Why appearances are insufficient: 32% mistook human-created content for AI-generated content in the Zefr + OM Media Trials study in the US and Canada. Read the study announcement ↗.
Why companies would adopt it
People want control. Violations can be expensive.
Earn consumer confidence
81% of US adults said they were concerned about how companies use their data in Pew’s 2023 survey. Give people a record they can inspect and a way to challenge misuse.
Avoid costly violations
Serious GDPR violations can bring fines up to €20 million or 4% of total annual global turnover, whichever is higher. Regulators can also restrict or ban processing.
Scraping has consequences
The Dutch regulator announced a €30.5 million fine against Clearview AI in September 2024 for illegal collection for facial recognition. Publicly accessible images do not remove privacy obligations.
The commercial proposal: buyers require evidence, platforms check current status, and suppliers with evidence that meets agreed requirements can earn access to business. The seal makes the record easy to open. It cannot make unlawful collection lawful or protect a company from enforcement.
Scraping is not automatically illegal. The data, legal basis, rights, purpose and jurisdiction determine the obligations. Survey concern does not prove that people will click a seal or that this proposal will increase sales; adoption and consumer use need testing.
AI company evidence directory / interactive proposal
A company map you can check, one record at a time.
A LUMAscape groups companies by their role in advertising. Our proposed version links each company to requirements, evidence and a scoped review result. Search a fictional advertiser, agency, data provider, AI tool or adtech partner.
Demonstration only. Every company, record and assessment below is fictional. Results apply only to the stated task or asset. The directory, checks and seal are proposals, not an adopted standard or legal certification.
From one action to the companies behind it
Check sources, permission and what happened.
Connect decisions, changes and handoffs.
Open the records within each review’s scope.
Follow evidence across agencies, tools and platforms.
Compare coverage, evidence gaps and corrections.
The same five fields connect each level: source, permitted use, responsibility, changes and verification. A passing record cannot certify an entire company.
Company + roleRequirementEvidence recordScoped reviewCorrection + recheck
Missing evidence means an unresolved assessment, not verified noncompliance. Every confirmed mismatch must point to the requirement and evidence behind it.
Fictional Advertiser / ORCHARD
Orchard Goods
Submitted checks metThe submitted evidence supports the stated use of this ad version. Other ads and company activities are outside this review.
Illustrative record and seal
Example Ad, Video 3
Ad example / AI assisted
Authorized
Replace the background; use the performer in US social ads through 31 Mar 2027.
Actual
Version 3 replaces the background. The performer’s face and voice are unchanged.
The seal opens the record, including its scope and review date. The symbol alone proves nothing.
Connected evidence / OG-003
Submitted evidence supports these checks
- Source
- Brand product photo and disclosed AI background tool recorded.
- Permitted use
- Performer agreement covers US social ads through 31 Mar 2027; withdrawal status checked at review.
- Responsibility
- Approver #4471: private identity, presence, authority and ability to reject verified for this version.
- Changes
- Background edit linked to version 3 and its content fingerprint.
- Verification
- Signed record, permission reference and edit history checked against the submitted version.
Proposed next step: Continue the assessed use. Recheck when the asset, permission or approved task changes.
Public view: requirement, decision, scope, version and review status. Authorized audit access: private identity and supporting evidence. People can challenge the record; the response is recorded.
The evidence we already expect
Burgers. Diamonds. Trading cards. Elevators.
What went into the grinder?
Beef can combine many inputs. An AI ad can combine a real photo, a synthetic voice, and licensed work. The ingredients need a history.
cyclonebill / Wikimedia Commons · source · CC BY-SA 2.0 · displayed at smaller size.
.jpg){kind=link}
Who checked it—and when?
The certificate records an inspection, its date, and validity. An ad receipt should make approval and authority equally easy to inspect.
NWRGeek / Wikimedia Commons · source · CC0 · historical 2012–2013 certificate.
Same carat. Same story?
Mined or lab-grown? What supports the origin and quality claims? Appearance alone cannot answer the questions behind a purchase.
Public-domain diamond photograph · Rawpixel / CC0 source.
Who issued it? Who owns it?
A digital card needs evidence of its issuer and ownership. A screenshot alone does not establish either.
NBA Collect by Topps · official product source · illustrative reference, not a claim of ownership.
Art and used cars make the same point: a convincing Picasso needs evidence, and “one careful owner” invites a request for service history. These analogies explain why we ask for records; they do not guarantee truth or safety.
The whole effort
From ethical principles to checkable behavior.
The Institute for Advertising Ethics Governance Lab identifies emerging risks and develops practical standards for accountable advertising, with IEEE Standards Association identified as a knowledge partner. The developing AI Advertising Trust Stack brings governance, technical evidence, and accountability into the same conversation.
Our Layer 4 contribution defines the checkable evidence needed across the data, content and decision lifecycle. It does not replace the other workstreams or speak on behalf of IAE or IEEE. The broader effort needs advertisers, creators, technologists, platforms, auditors, and people affected by advertising to help define what justified trust requires.
Define intent.
Establish purpose and boundaries before testing the system’s behavior.
Define controls.
Set approvals, operating limits, required check-ins and escalation.
Supply evidence.
Preserve sources, permitted use, responsibility, changes and repeatable verification.
Layers 5 and 6 use the evidence for accountability and ongoing assurance. Layer 2 addresses the workforce. This page focuses on the evidence handoffs rather than defining every workstream.
These are the adjacent workstreams relevant to this example, not a complete specification of the entire Trust Stack. Read IAE’s official Governance Lab overview ↗.
Progress update · 11 October 2026
What we have—and what we still need.
Current status: working proposal and illustrative prototypes.
This is developing work, not an adopted IEEE standard, certification scheme, or completed end-to-end system.
A practical evidence record.
- Five questions covering origin, permitted use, responsibility, changes, and checking.
- A fictional Example Ad receipt and fair-review scenarios.
- Record marks on social ads, video, and editorial content.
- A public source register crediting contributions and identifying gaps.
- A six-slide presentation for ten minutes, with a matching file-fingerprint check.
The controls behind the claim.
- What proof is sufficient for each risk, and where the record starts.
- Who verifies, how long evidence is kept and how a person challenges a record.
- Authority, private identity, and proof-of-presence controls.
- Signed approval and C2PA integration across ad-tech handoffs.
- Who answers for missing evidence at a company handoff, and how review stays affordable.
Prepared for The Intention Advantage at Omnicom on 12 October 2026. The current presentation incorporates the Summit I report, defines the ad-agent and data scope, and distinguishes signed Content Credentials from watermarks. The committee still needs to agree the evidence requirements and test them in a real workflow.
Progress updates on this page are dated and added as the work develops. No automated reporting or mailing-list subscription is implied.
Layer 4 · five questions
The receipt should answer ordinary questions.
| Question | What the evidence should show | How someone uses it |
|---|---|---|
| Where did it come from? | Audience-data supplier, creator, tools, AI use, source records and gaps. | Check origin, credit, and ownership claims. |
| What can it be used for? | Applicable permission or license, purpose, region, limits, expiry and withdrawal. | Test a disputed use against permission. |
| Who is responsible? | Responsible company, human or automated decision, approval authority and ability to reject or stop. | Find who controlled the step and hear their response. |
| What changed? | Content edits, changes in agent methods, versions and intervention history. | Compare later actions or versions with their approved purpose and limits. |
| How can it be checked? | Signatures where available, repeatable checks, action logs and protected identity evidence. | Audit without publishing private identities. |
Provenance records origin and changes. A valid record still needs to be checked against the applicable requirements. Human presence must come with authority and the power to reject or stop; private identity evidence stays under controlled access.
Small mark · deeper evidence
See how the record could appear in real content.
Illustrative placements: the mark opens the record for review. These examples do not carry verified approval.
Example Brand · Sponsored
Example Ad / AI assisted
Open content record ↗Example Ad, Video 3
Illustrative frame / AI assisted
Open content record ↗What went into this ad?
The source and edits.
The permission to use it.
The person or company responsible.
Illustrative article / AI assisted
Open content record ↗Our everyday examples start with diamonds, digital trading cards, art, used cars, beef and elevator certificates. We already ask for origin, permission, history, and inspection evidence elsewhere. Why should data or AI be different?
Contributions and boundaries
Different expertise. A shared record.
Jay Mandel
Narrative, permission and accountability questions, stakeholder use, and the connection between evidence and decisions.
Sean Meier
Layer 4 technical collaboration: source collection, detection signals, privacy-preserving approval checks, and record carriage.
Operational alignment
Compare authorized purpose, permissions and disclosures with actual behavior. Preserve the evidence that workflow and accountability owners need to investigate and correct a mismatch.
C2PA Content Credentials provide signed content-history records. Watermarks are embedded signals that may help recover those records; AI-detection results are estimates. These components do not establish permission or legal compliance on their own. The full evidence record and its governance remain proposals. ChatGPT assisted with this page, deck, and prototype implementation. The source register distinguishes actual production facts, attributed discussion, research references, and fictional examples.
Read the contributor and source table ↗Proposal for discussion
A LUMAscape that rewards evidence.
A LUMAscape groups advertising companies by role. Our proposed evidence map connects brands, agencies, data providers, AI tools and adtech partners to a requirement, its supporting records, reviewer, scope, date and result. People and rights holders can inspect the public record and challenge disputed use.
The proposal is to value checkable evidence in purchasing, distribution, and risk review. That is not a promise of lower insurance premiums, guaranteed reach, or an existing certification.
Explore the ecosystem and incentives ↗Follow the AI Advertising Trust Stack work.
Read the progress. Bring a real question.
What proof is sufficient for the risk? Who verifies it and retains it? Who answers when the evidence stops at a company handoff?