Information required to perform an audit
To conduct a semantic integrity audit, AI ScanLab requires the following from the client.
1. System overview
- Description of AI systems involved
- Identification of models, platforms, or agents
- Explanation of how AI outputs are used operationally
2. Access context
No administrative access is required.
Clients may provide:
- Public URLs
- Test environments
- Sandbox accounts
- Non-privileged user access
Audits are performed as a simulated external or internal user.
3. Scope definition
Clients must define:
- The content, products, or narratives to be audited
- The relevant languages or regions
- The time horizon (single snapshot or longitudinal)
4. Risk context
Understanding risk is critical.
Clients should indicate whether the audit relates to:
- Revenue impact
- Regulatory exposure
- Brand or reputational risk
- Operational decision-making
5. Legal and compliance constraints
Any constraints related to:
- Data protection
- Confidentiality
- Regulatory frameworks
must be disclosed before the audit begins.
What is not required
- Infrastructure credentials
- Source code
- Training data
- Model weights
- Internal prompts
