Back to the MIT repository
7. AI System Safety, Failures, & Limitations1 - Pre-deployment

Value Chain and Component Integration

Non-transparent or untraceable integration of upstream third-party components, including data that has been improperly obtained or not processed and cleaned due to increased automation from GAI; improper supplier vetting across the AI lifecycle; or other issues that diminish transparency or accountability for downstream users.

Source: MIT AI Risk Repositorymit767

ENTITY

1 - Human

INTENT

2 - Unintentional

TIMING

1 - Pre-deployment

Risk ID

mit767

Domain lineage

7. AI System Safety, Failures, & Limitations

375 mapped risks

7.4 > Lack of transparency or interpretability

Mitigation strategy

1. Institute a mandatory, comprehensive AI Supplier Risk Management (SRM) and due diligence framework. This framework must require contractual agreements that explicitly define security standards, data provenance expectations, and audit rights, as well as pre-assessment of third-party models for adherence to fairness and explainability criteria prior to integration. 2. Establish and maintain an immutable, end-to-end data and component lineage system. This system must document all data acquisition (including proper consent/licensing validation), preprocessing steps (e.g., cleaning, de-identification), and the specific versioning of all integrated third-party code and model components, ensuring full traceability from source to final system output. 3. Execute rigorous pre-deployment technical validation protocols, including independent third-party audits and adversarial testing (e.g., 'red teaming'). This activity is critical to verify that the integrated system's behavior aligns with safety and compliance objectives, specifically confirming the mitigation of latent bias or data-related quality issues amplified by automated GAI processes.