Mapping NIST’s framework requirements on the HAIP Guides

Characteristics of trustworthy AI	Requirement	HAIP Guides
Safe	Protection of human life, health, property, or the environment from being endangered.	Evaluate AI product safety prior to deployment Test safety and security Identify and mitigate risks Prevent inappropriate use of the AI product Monitor AI performance
Secure/ resilient	Ability to withstand unexpected adverse events or unexpected changes in their environment or use	Test safety and security Prevent inappropriate use of the AI product Determine if updating or decommissioning is necessary Minimize disruptions from decommissioning
Explainable and interpretable	Representation of the mechanisms underlying AI systems’ operation Meaning of AI systems’ output in the context of their designed functional purposes	Disseminate information to clinical end-users Establish training on how to interpret AI-generated recommendations
Privacy-enhanced	Human autonomy, identity, and dignity are safeguarded against intrusion or disclosure of facets of people’s identities.	Understand which regulatory schemes apply to the AI product Assess legal risks, including privacy laws
Fair and manage bias	Concerns for equality and equity are addressed by mitigating issues such as harmful bias and discrimination.	Address bias at initial stages in problem prioritization Include ethics experts Assess ethics and equity of outside products Define performance targets Perform a risk analysis with attention to sources of harm Monitor biases and ethical concerns Monitor bias from data shifts and product use
Accountable and transparent	The extent to which information about an AI system and its outputs is available to individuals interacting with such a system. The extent to which responsibility for aspects of AI systems are clearly identified.	Audit investment decisions Share performance metrics with all stakeholders to ensure transparent communication Disseminate information to clinician end-users Audit AI solutions and the work environment Disseminate information about updates to clinician end-users
Valid and reliable	Confirmation, through the provision of objective evidence, that the requirements for a specific intended use or application have been fulfilled. The AI system can perform as required, without failure, for a given time interval, under given conditions.	Local validation Sustain improved outcomes Evaluate expansion to new settings Minimize disruptions from decommissioning an Al product