What Is an AI Transparency Evidence Pack? Documenting Company Use of AI

Updated: 3 September 2026 · Category: AI Transparency

What Is an AI Transparency Evidence Pack? Documenting Company Use of AI

An AI Transparency Evidence Pack is an audit evidence package linking the model used, purpose, prompt/brief, generation time, human review, rights checks, disclosure decision and published version.

Contents

Short answer
Why does it matter?
What should companies do?
How should it be measured?
Sources
Frequently asked questions

Short Answer

An AI Transparency Evidence Pack is an audit evidence package linking the model used, purpose, prompt/brief, generation time, human review, rights checks, disclosure decision and published version.

Why Does It Matter?

Saying “we used AI” is not audit evidence on its own. A company must be able to show afterwards which model was used for which purpose in a campaign, where a human made decisions, which output was published and how the disclosure decision was reached.

The AI Transparency Evidence Pack is not a mandatory document defined by that name in the AI Act. It is B10’s documentation model for turning Article 50 transparency obligations and corporate AI governance needs into operational evidence. For teams producing many creative assets in particular, the aim is to answer “which version was published and why?” through a single file.

What should companies do? Eight evidence layers to include in the Evidence Pack

Use-case record: The business purpose and process for which the AI system is used.
Model/provider information: The name and, where possible, version of the service, model or generation tool used.
Input source: The prompt/brief, source of uploaded data or reference material, and sensitive data checks.
Output record: Distinguish the main generated versions from the final file actually published.
Human oversight: Verification and changes made by the team or person reviewing the content.
Rights checks: Checks on trademarks, copyright, personality rights, and use of images and audio.
Transparency decision: Assessment of whether labelling/disclosure is required under Article 50 or relevant advertising rules.
Publication and change history: Channel, date, final URL/creative asset, and records of subsequent corrections or removals.

Proposed measurement model for B10

Evidence field	Required internal record	Control
Use-case	Purpose + process owner	Complete / incomplete
Model	Provider + model/version	Traceability
Input	Brief/prompt + data source	Data/rights risk
Human review	Checker + date	Review coverage
Disclosure	Decision + rationale + label applied	Transparency completeness
Final Output	Published file/URL	Version match

The practical goal is not to produce hundreds of pages for every AI-assisted publication. Evidence depth should increase with risk. A simple image variation and a deepfake advertisement impersonating a real person should not be managed under the same evidence standard.

Minimum and extended record levels can be defined for the Evidence Pack. For a low-risk text draft, the use case, tool used, human review and final version record may suffice. A campaign involving a real person’s image, customer data, health or financial claims, deepfakes, high-reach advertising or regulatory sensitivity should retain more detailed source data, rights checks, disclosure rationale and approval chains.

A key feature of the evidence package is linking the final output to the decision record. A team may generate dozens of AI variants; an audit needs to establish which file was actually published and which checks were performed on it. Technical references such as creative asset IDs, version numbers, file hashes or publication URLs can therefore be included in the Evidence Pack.

The Evidence Pack also supports change management. If a product claim, image, disclosure or targeting rule changes after publication, retain the history of both old and new versions. This lets the company show the current content’s status and how decisions changed over time.

Responsibility for the Evidence Pack should be defined in advance. Marketing can maintain creative assets and briefs, technical teams the systems and versions used, and legal or compliance teams the disclosure and rights assessments. These parts should be linked under a common case ID rather than left in separate systems. During an audit or dispute, the company can then work from one traceable record instead of retrospectively searching email, file servers and campaign tools for evidence.

Expert Assessment

The AI Transparency Evidence Pack is not a standard mandatory document named in legislation; it is a governance model for systematically retaining decisions and evidence about a company’s AI use.
The value lies in traceability, not file counts: which AI use served which purpose, which model or tool was used, who performed human review and how the required disclosure decision was made.
Correct versioning and evidence links greatly reduce the need to search retrospectively across email, file servers and campaign tools during audits or internal checks.

B10 Digital Agency Digital Risk and Compliance Analysis

B10 can design an internal AI Transparency Evidence Pack linking the AI use inventory, model and tool records, human approval, disclosure decisions and version history.

Request an Analysis