Back to the MIT repository
7. AI System Safety, Failures, & Limitations2 - Post-deployment

Non-disclosure

Content might not be clearly disclosed as AI generated.

Source: MIT AI Risk Repositorymit1298

ENTITY

1 - Human

INTENT

3 - Other

TIMING

2 - Post-deployment

Risk ID

mit1298

Domain lineage

7. AI System Safety, Failures, & Limitations

375 mapped risks

7.4 > Lack of transparency or interpretability

Mitigation strategy

1. Mandate Explicit and Unambiguous Disclosure Mechanisms Implement technical and policy-based requirements for prominently watermarking, labeling, or otherwise clearly flagging all AI-generated output. This includes establishing a standard for the location and format of the disclosure (e.g., metadata, disclaimers, or signature lines) to ensure the recipient is immediately and unequivocally informed of the content's synthetic provenance. 2. Establish a Comprehensive AI Transparency and Governance Framework Develop an internal framework that standardizes disclosure protocols, assigns accountability for compliance, and defines the necessary degree of transparency for different risk levels of content. The framework should require regular audits to verify that AI usage and disclosure policies are consistently adhered to by all system developers and end-users. 3. Implement Contextual and Granular Transparency Mechanisms Move beyond simple disclosure by offering accessible explanations (outcome and process transparency) that detail the extent of AI involvement. This should include communicating the specific AI model utilized, the level of human oversight or modification applied, and the system's intended purpose, thereby providing the user with essential context for evaluating the content's validity and reliability.