Towards a Standard for Identifying and Managing Bias in Artificial Intelligence
This NIST special publication outlines preliminary socio-technical guidance for identifying, understanding, managing, and reducing bias in artificial intelligence systems. It examines three categories of AI bias (systemic, statistical, and human) and discusses challenges across datasets, test, evaluation, validation, and verification (TEVV), and human factors. The document serves as a foundational step toward more detailed guidance to improve public trust and promote trustworthy and responsible AI.
Summary and content fields are drawn from the document text and are pending editorial review.
Who issued it
Who it governs
Individuals and groups responsible for designing, developing, deploying, evaluating, and governing AI systems.
What the document contains
The document addresses the sources and categories of AI bias spanning datasets, testing and evaluation, and human factors. It provides preliminary guidance on socio-technical approaches, human-centered design, governance, and organizational practices to mitigate AI bias.
The document explicitly states that it does not establish any legal standard, law, or mandatory regulation, nor does it serve as a complete final standard.
- AI bias cannot be fully understood or mitigated through purely computational and mathematical methods alone.
- Three primary categories of AI bias exist: systemic, statistical and computational, and human.
- Datasets frequently suffer from representation issues, historical inequities, and lack of socio-technical contextualization.
- Test, evaluation, validation, and verification (TEVV) face challenges such as lack of ground truth, unobservable constructs, and epistemic uncertainty.
- Human-in-the-loop and expert-driven configurations are not infallible guarantees against bias and require organizational governance and cultural support.
1. Purpose and Scope 2. AI Bias: Context and Terminology 3. AI Bias: Challenges and Guidance 4. Conclusions 5. Glossary
How it scores
Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.
Accountability
4/5Clear responsibility for AI outcomes
Ensuring that a specific team, and often, a specific individual – such as a Chief Model Risk Officer, as is now common in large consumer finance organizations – is responsible for bias management in AI systems is a fundamental accountability mechanism.
3.4.1 Governance Guidance
screening score · 2026-10-23
Fairness
4/5Non-discrimination and equitable treatment
Specific methods for algorithmic mitigation of bias for many different machine learning tasks have been delineated or surveyed in recent studies
3.2.2 TEVV Guidance
screening score · 2026-10-23
Human Dignity
2/5Respect for the inherent worth of persons
A human-centered future will also support human values, respect human dignity, and raise appreciation for the human capacities that bring creative discoveries and inventions.
3.3.2 Human Factors Guidance
screening score · 2026-10-23
Human Oversight
3/5Humans remain in control of consequential decisions
Human factors such as participatory design techniques and multi-stakeholder approaches, and a human-in-the-loop are also important for mitigating risks related to AI bias.
Executive Summary
screening score · 2026-10-23
Privacy
2/5Protection of personal data and private life
This mainly stems from the perceptions and fears about loss of control and privacy
1. Purpose and Scope
screening score · 2026-10-23
Transparency
3/5Openness about how AI systems work and are used
The importance of transparency, datasets, and test, evaluation, validation, and verification (TEVV) cannot be overstated.
Executive Summary
screening score · 2026-10-23
Enforcement
2/5Consequences and compliance mechanisms with real force
Availability of feedback channels allow system end users to flag incorrect or potentially harmful results, and seek recourse for errors or harms.
3.4.1 Governance Guidance
screening score · 2026-10-23