Skip to content

1.4.2. Application of three trustworthy categories (ethical, legal, and stable) at each stage of the life cycle

Trustworthiness Across the AI Lifecycle

Each stage of the AI development lifecycle provides an opportunity to implement the trustworthy categories of ethical, legal, and stable practices. By understanding how these categories are applied, we can see that trust in AI systems extends beyond technical reliability to meeting broader societal expectations.

The table below outlines how these trustworthiness categories can be practically embedded into each phase of the AI lifecycle.

Table 3: Applying Trustworthiness Categories Across AI Lifecycle Stages

Life Cycle Stages Ethical Trustworthiness Legal trustworthiness Stability
Planning Phase Human-Centered and Fair Goals Review.

UN Guidelines: Designing to Prevent Worsening Social Inequalities		 Clear data usage plan considering data regulations such as GDPR Identify anticipated technical risks and create a mitigation plan
Data collection and management phases Get representative data from diverse populations. Eliminate bias with the IBM AI Fairness 360 toolkit Privacy Regulation Compliance.

Ensuring
 Data Traceability and Explainability (EU AI Act)		 Maintain data integrity.

Prevent
 loss and tampering Technical Measures
Model Design Phase Design mechanisms to ensure algorithmic fairness.

Analyze model bias and simulate fairness with the Google What-If Tool		 Meet regulatory requirements.

Prepare design documentation Gain
 external verifiability		 Robust against hostile attacks, securing stability in high-risk areas such as finance
Evaluation and Validation Phase Test performance in a variety of environments
to ensure that it is not disadvantageous to a particular population		 Document test procedures and archive results for compliance with regulatory requirements

Validate response to unexpected input/environment changes
Deployment and monitoring phases Reflecting user feedback and continuing to evaluate ethical issues EU AI Act: Post-Deployment Performance/Safety

Monitoring and Addressing Emerging Risks		 Real-time performance tracking.

Utilize
anomaly detection monitoring tools (e.g., financial AI fraud transaction detection)

As shown, each lifecycle phase plays a distinct role in implementing ethical, legal, and stability safeguards. When applied systematically, these categories help AI systems go beyond technical goals and build trustworthiness in both function and public perception.

TRAI Challenges

Embedding Trust in the AI Lifecycle

Scenario:
Your team is designing an AI system for predicting job applicant success. You are in the early planning stages and need to create a trust map that outlines potential risks and safeguards across the AI lifecycle.

Task:
Using the five standard lifecycle stages below, do the following:
1. Identify one risk to trustworthy AI at each stage.
2. Classify the risk by trust pillar (Ethical, Legal, Stable).
3. Suggest a strategy or control to mitigate that risk.

Instructions:
- Complete the table below with clear, concise entries (1–2 sentences per cell).
- Base your answers on the concepts and case studies discussed earlier in the chapter.
- You may complete this table on a separate worksheet or digital notebook to reflect on how each lifecycle stage can protect against harm.

Lifecycle Stage Identified Risk Trust Pillar (Ethical/Legal/Stable) Mitigation Strategy
Planning and Design
Data Collection & Preprocessing
Model Training
Deployment and Integration
Monitoring and Feedback