IA-02 Transparency

2 sub-dimensions · 12 risk scenarios

Ensures that activities, decisions, capabilities, and limitations regarding the AI system are disclosed to all relevant stakeholders in a comprehensive, clear, and understandable manner. Specific evaluations include mandatory disclosure of AI interaction and generated content (to prevent deepfake deception) , communication of system capabilities and limitations (to avoid automation bias) , and transparency of training data sources and basic model architecture.

🎯 BenefitsProactively discloses AI identity, system capabilities, and operational limitations to establish reasonable user expectations, guide responsible collaboration, and enhance public trust.

⚠️ Harms / RisksFailing to label AI identity causes users to mistake it for a real person and leak sensitive personal data, or failing to disclose system limitations causes users to develop "automation bias" and overly rely on blind decision-making.

Insufficient Information on System Capabilities7 risk scenarios · ISO 23894 A.12

DefinitionEvaluates whether interested parties can obtain sufficient information to understand the AI system's capabilities, constraints, and limitations.

🎯 BenefitsFully discloses the AI system's capability boundaries, applicable contexts, constraints, and known weaknesses, guiding users toward reasonable expectations and preventing automation bias.

⚠️ Harms / RisksSuppliers or development teams fail to adequately communicate system limitations and unsuitable use cases, leading users to place blind trust in the AI or apply it in untested, hazardous environments.

Representative ScenariosS0013 Supplier Opacity Preventing Risk Assessment S0014 Third-Party Service Provider Risk S0058 Misleading Output from Explanation Tools S0060 Knowledge Gaps Leading to Blind Decisions S0101 Insufficient Transparency Undermining Oversight S0102 Version Changes Not Communicated to Users S0103 Opaque Source Attribution

Undisclosed AI Interaction5 risk scenarios · ISO 23894 A.12

DefinitionEvaluates whether interested parties are aware that they are interacting with an AI system.

🎯 BenefitsMandates disclosure of AI interaction identity and the provenance of generated content, guarding against anthropomorphic deception, deepfake-based misdirection, and commercial fraud.

⚠️ Harms / RisksCustomer service or conversational systems interact in an anthropomorphized manner without disclosing their AI identity, causing users to divulge highly sensitive information under false assumptions or fall victim to deepfake scams.

Representative ScenariosS0012 No Planned AI Disclosure Mechanism S0152 Generated Content Reinforcing Stereotypes S0165 Anthropomorphism Producing Misplaced Trust S0200 Blurred Authenticity and Cognitive Disorientation S0201 Content Homogenization and Declining Diversity