AI Transparency — Disclosure and Documentation Requirements

Definition

AI transparency is the obligation and practice of providing clear, accurate, and accessible information about AI systems — their purpose, capabilities, limitations, decision logic, data sources, and potential risks — to the relevant stakeholders: operators deploying the system, individuals affected by its decisions, regulators, auditors, and the public. It encompasses both technical transparency (documentation of system architecture, training data, and performance metrics) and functional transparency (human-understandable explanations of how the system works and how decisions are made).

ISO TR 24028 provides international guidance on transparency in AI, covering transparency requirements for different stakeholder types and deployment contexts. The EU AI Act mandates specific transparency obligations at multiple levels: technical documentation and logging requirements for high-risk systems, transparency to deployers about system capabilities and limitations, disclosure requirements for AI systems interacting with individuals, and transparency obligations for general-purpose AI model providers.

Why it matters operationally

Transparency is the enabling condition for accountability, oversight, and trust. Without transparency about how AI systems work and why they produce specific outputs, human oversight becomes impossible — reviewers cannot meaningfully supervise decisions they cannot understand. Individuals affected by AI decisions cannot challenge them. Regulators cannot assess compliance. Courts cannot evaluate liability.

The regulatory pressure on AI transparency is increasing across all major jurisdictions. The EU AI Act mandates technical documentation, logging, and disclosure requirements. State-level AI laws in the US are introducing algorithmic transparency obligations. Financial regulators are requiring model risk documentation. Healthcare regulators are demanding explainability for AI-assisted clinical decisions. Transparency is moving from a governance principle to an enforceable legal obligation.

Regulatory framework

Framework	Transparency obligations
EU AI Act	Technical documentation per Annex IV for high-risk systems. Information obligations to deployers. Disclosure of AI nature when interacting with individuals. Documentation of capabilities and limitations for GPAI models.
ISO TR 24028	Technical guidance on transparency in AI for different stakeholder groups.
ISO/IEC 42001	Annex A includes transparency and documentation controls as part of the management system.
GDPR — Arts. 13, 14	Information about AI use in personal data processing, including logic of automated decisions.

How Zertia evaluates it

Zertia evaluates transparency through the AI Model Audit — specifically examining technical documentation completeness, logging adequacy, and whether transparency mechanisms are proportionate to the deployment context and the stakeholders affected. The Ethical AI Mark evaluates conformity with ISO TR 24028 (Transparency) as one of its four international ethics assessment dimensions.

[AI Model Audit] · Ethical AI Mark