Skip to content
AIGP Quest

Glossary

Trustworthy AI

Written with AI from the sources below and checked against our facts sheet. Last checked . How this site is made · Spot an error?

Part of: NIST AI Risk Management Framework

Trustworthy AI is AI that people can reasonably rely on because its key qualities are designed in, tested and evidenced. NIST lists seven characteristics: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. The EU's 2019 ethics guidelines frame it as lawful, ethical and robust.

In plain English

You trust a bridge because engineers designed it to standards, inspectors check it, and someone answers if it fails. You do not need to see the math. Trustworthy AI works the same way: trust is earned through properties that can be tested, and evidence that they hold.

The NIST AI RMF names seven characteristics:

  1. Valid and reliable
  2. Safe
  3. Secure and resilient
  4. Accountable and transparent
  5. Explainable and interpretable
  6. Privacy-enhanced
  7. Fair, with harmful bias managed

NIST shows valid and reliable as the base for the others. Accountable and transparent cuts across all of them. NIST also warns that trustworthiness is "only as strong as its weakest characteristics."

Example

Marrowby County, a fictional local government, is buying an AI tool to flag benefit applications for extra review. Its procurement scorecard asks vendors for evidence on each characteristic:

CharacteristicEvidence Marrowby asks for
Valid and reliableAccuracy on county data, not just the vendor's benchmark
SafeWhat happens when the tool fails, and a manual fallback
Secure and resilientPenetration test and red team results
Accountable and transparentNamed owners, and notice to applicants that AI is used
Explainable and interpretableReasons caseworkers can read for each flag
Privacy-enhancedData minimization and retention terms
Fair, with harmful bias managedError rates by group, using a fairness metric chosen in advance

One vendor scores highest on accuracy but cannot explain individual flags. The county weighs that trade-off against its duty to explain decisions. It picks a slightly less accurate model that caseworkers can interpret.

Why it matters

  • NIST. Trustworthiness is the organizing idea of the AI RMF, and its MEASURE function evaluates systems against these characteristics. NIST says they involve trade-offs, such as predictive accuracy against interpretability, and must be balanced in context.
  • EU. The High-Level Expert Group's Ethics Guidelines for Trustworthy AI (8 April 2019) define trustworthy AI as lawful, ethical and robust, with seven key requirements. The EU AI Act aims to promote "human-centric and trustworthy" AI (Art. 1(1)). Recital 27 recalls the guidelines' seven principles.
  • OECD. The OECD AI Principles, adopted in 2019 and updated in May 2024, present themselves as principles for trustworthy AI.

How the AIGP exam tests it

Lesson II.D.2 teaches NIST's seven characteristics. Know them cold.

  • Spot the impostor. "Profitable," "scalable," "cost-effective" and "fully autonomous" are never on the list.
  • Structure. Valid and reliable is the base. Accountable and transparent cuts across the rest.
  • Trade-offs. NIST does not rank the characteristics. The best answer balances them in context and documents the choice.
  • What, how, why. In NIST's terms, transparency answers what happened, explainability answers how, and interpretability answers why. See explainable AI.
  • NIST vs EU lists. The EU guidelines also have seven requirements, with different labels. "Human agency and oversight" is on the EU list, not NIST's.

Frequently asked questions

What are the 7 characteristics of trustworthy AI?

NIST's seven are valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. Valid and reliable is the base. Accountable and transparent relates to all the others. NIST does not rank them, and organizations balance them in context.

What are the EU requirements for trustworthy AI?

The 2019 Ethics Guidelines for Trustworthy AI say AI should be lawful, ethical and robust. The first three of seven key requirements are human agency and oversight, technical robustness and safety, and privacy and data governance. The others are transparency; diversity, non-discrimination and fairness; societal and environmental well-being; and accountability. The guidelines are voluntary, but Recital 27 of the EU AI Act refers to them.

What is the difference between trustworthy AI and responsible AI?

Trustworthy AI describes qualities of the system, the properties that justify relying on it. Responsible AI describes how an organization behaves when it designs, builds and uses AI. The two overlap heavily. A responsible AI program is how an organization produces systems that are trustworthy.

Can AI be trustworthy if it is not explainable?

Partly. NIST treats explainability as one of seven characteristics that involve trade-offs, so some contexts accept less of it. But NIST also says trustworthiness is only as strong as its weakest characteristic. Where decisions affect people's rights, laws such as the GDPR and the EU AI Act expect meaningful explanations.

Related

Sources

  1. NIST AI 100-1, AI Risk Management Framework 1.0 (January 2023), section 3
  2. European Commission, Ethics Guidelines for Trustworthy AI (High-Level Expert Group on AI, 8 April 2019)
  3. EU AI Act Service Desk, Recital 27
  4. Regulation (EU) 2024/1689 (EU AI Act), Art. 1(1), EUR-Lex
  5. OECD AI Principles (2019, updated May 2024)
Spot an error? Tell us