2/25/25, 7:21 PM Playbook

Playbook
Type Title AI Actors Topics Description

Legal and regulatory

Legal and Regulatory,
Govern Governance and requirements involving AI are
Govern Governance, AI Actor
1.1 Oversight understood, managed, and
Training
documented.

Trustworthy
Characteristics,
Governance, Validity
and Reliability, Safety,
The characteristics of
Secure and Resilient,
Govern Governance and trustworthy AI are integrated
Govern Accountability and
1.2 Oversight into organizational policies,
Transparency,
processes, and procedures.
Explainability and
Interpretability,
Privacy, Fairness and
Bias

Processes and procedures

are in place to determine the
Govern Governance and Risk Tolerance, needed level of risk
Govern
1.3 Oversight Governance management activities based
on the organization's risk
tolerance.

The risk management

process and its outcomes are
Risk Management, established through
Govern Governance and
Govern Governance, transparent policies,
1.4 Oversight
Documentation procedures, and other
controls based on
organizational risk priorities.

Ongoing monitoring and

periodic review of the risk
management process and its
Governance and Monitoring,
outcomes are planned,
Govern Oversight, Governance,
Govern organizational roles and
1.5 Operation and Continual
responsibilities are clearly
Monitoring Improvement
defined, including
determining the frequency of
periodic review.

Govern Govern Governance and Risk Management, Mechanisms are in place to

1.6 Oversight Governance, Data, inventory AI systems and are
about:blank 1/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

Documentation resourced according to
organizational risk priorities.

Processes and procedures

are in place for
decommissioning and
AI Deployment,
Govern Decommission, phasing out of AI systems
Govern Operation and
1.7 Governance safely and in a manner that
Monitoring
does not increase risks or
decrease the organization’s
trustworthiness.

Roles and responsibilities and

lines of communication
related to mapping,
Govern Governance and Governance, Risk
Govern measuring, and managing AI
2.1 Oversight Culture
risks are documented and are
clear to individuals and teams
throughout the organization.

The organization’s personnel

and partners receive AI risk
management training to
Govern Governance and Governance, AI Actor enable them to perform their
Govern
2.2 Oversight Training duties and responsibilities
consistent with related
policies, procedures, and
agreements.

Executive leadership of the

organization takes
Govern Governance and Governance, Risk responsibility for decisions
Govern
2.3 Oversight Tolerance about risks associated with AI
system development and
deployment.

Decision-makings related to
mapping, measuring, and
managing AI risks throughout
Governance and Diversity,
Govern the lifecycle is informed by a
Govern Oversight, AI Interdisciplinarity,
3.1 diverse team (e.g., diversity
Design Governance
of demographics, disciplines,
experience, expertise, and
backgrounds).

Govern Govern AI Design Human-AI teaming, Policies and procedures are

3.2 Human oversight, in place to define and
Governance, AI Actor differentiate roles and
Training responsibilities for human-AI

about:blank 2/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

configurations and oversight
of AI systems.

Organizational policies, and

practices are in place to
AI Design, AI
foster a critical thinking and
Development, AI Risk Culture,
Govern safety-first mindset in the
Govern Deployment, Governance,
4.1 design, development,
Operation and Adversarial
deployment, and uses of AI
Monitoring
systems to minimize negative
impacts.

Organizational teams
AI Design, AI document the risks and
Development, AI Risk Culture, potential impacts of the AI
Govern
Govern Deployment, Governance, Impact technology they design,
4.2
Operation and Assessment develop, deploy, evaluate and
Monitoring use, and communicate about
the impacts more broadly.

TEVV, Operation
Risk Culture,
and Monitoring, Organizational practices are
Governance, AI
Govern Governance and in place to enable AI testing,
Govern Incidents, Impact
4.3 Oversight, identification of incidents,
Assessment, Drift,
Fairness and and information sharing.
Fairness and Bias
Bias

Organizational policies and

AI Design, practices are in place to
Governance and collect, consider, prioritize,
Oversight, AI and integrate feedback from
Participation,
Govern Impact those external to the team
Govern Governance, Impact
5.1 Assessment, that developed or deployed
Assessment
Affected the AI system regarding the
Individuals and potential individual and
Communities societal impacts related to AI
risks.

AI Impact Mechanisms are established

Assessment, to enable AI actors to
Participation,
Govern Governance and regularly incorporate
Govern Governance, Impact
5.2 Oversight, adjudicated feedback from
Assessment
Operation and relevant AI actors into system
Monitoring design and implementation.

about:blank 3/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

Policies and procedures are

Third-party in place that address AI risks
Third-party, Legal and
entities, associated with third-party
Govern Regulatory,
Govern Operation and entities, including risks of
6.1 Procurement, Supply
Monitoring, infringement of a third party’s
Chain, Governance
Procurement intellectual property or other
rights.

AI Deployment, Contingency processes are in

Third-party,
TEVV, Operation place to handle failures or
Govern Governance, Risk
Govern and Monitoring, incidents in third-party data
6.2 Management, Supply
Third-party or AI systems deemed to be
Chain
entities high-risk.

A determination is made as to
AI Deployment,
whether the AI system
Operation and
Manage AI Deployment, Risk achieves its intended purpose
Manage Monitoring, AI
1.1 Assessment and stated objectives and
Impact
whether its development or
Assessment
deployment should proceed.

AI Deployment, Treatment of documented AI

Operation and risks is prioritized based on
Manage
Manage Monitoring, AI Risk Tolerance impact, likelihood, or
1.2
Impact available resources or
Assessment methods.

Responses to the AI risks

deemed high priority as
AI Deployment, identified by the Map
Operation and function, are developed,
Manage Legal and Regulatory,
Manage Monitoring, AI planned, and documented.
1.3 Risk Tolerance
Impact Risk response options can
Assessment include mitigating,
transferring, avoiding, or
accepting.

Negative residual risks

AI Deployment,
(defined as the sum of all
Operation and
Manage unmitigated risks) to both
Manage Monitoring, AI Risk Response
1.4 downstream acquirers of AI
Impact
systems and end users are
Assessment
documented.

Manage Manage AI Deployment, Risk Tolerance, Trade- Resources required to

2.1 Operation and offs manage AI risks are taken
Monitoring, AI into account, along with
Impact viable non-AI alternative
Assessment, systems, approaches, or
about:blank 4/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

Governance and methods – to reduce the
Oversight magnitude or likelihood of
potential impacts.

AI Deployment,
Operation and
Monitoring, AI Mechanisms are in place and
Manage AI Deployment, Drift,
Manage Impact applied to sustain the value
2.2 Societal Values
Assessment, of deployed AI systems.
Governance and
Oversight

Procedures are followed to

AI Deployment,
Manage respond to and recover from
Manage Operation and Risk Response
2.3 a previously unknown risk
Monitoring
when it is identified.

Mechanisms are in place and

applied, responsibilities are
AI Deployment,
assigned and understood to
Operation and Risk Response,
Manage supersede, disengage, or
Manage Monitoring, Decommission, Risky
2.4 deactivate AI systems that
Governance and Emergent Behavior
demonstrate performance or
Oversight
outcomes inconsistent with
intended use.

Third-party AI risks and benefits from

entities, third-party resources are
Manage Third-party, Supply
Manage Operation and regularly monitored, and risk
3.1 Chain
Monitoring, AI controls are applied and
Deployment documented.

Third-party Pre-trained models which are

entities, used for development are
Manage Pre-trained models,
Manage Operation and monitored as part of AI
3.2 Monitoring
Monitoring, AI system regular monitoring
Deployment and maintenance.

Post-deployment AI system
AI Deployment,
monitoring plans are
Operation and Monitoring,
implemented, including
Monitoring, Participation, AI
mechanisms for capturing
End-Users, Deployment, AI
Manage and evaluating input from
Manage Human Factors, Incidents, Risk
4.1 users and other relevant AI
Domain Experts, Response,
actors, appeal and override,
Affected Adversarial, Risky
decommissioning, incident
Individuals and Emergent Behavior
response, recovery, and
Communities
change management.

about:blank 5/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

TEVV, AI Design,
AI Development, Measurable activities for
AI Deployment, Monitoring, Impact continual improvements are
Operation and Assessment, Risk integrated into AI system
Manage
Manage Monitoring, Assessment, updates and include regular
4.2
End-Users, Continual engagement with interested
Affected Improvement parties, including relevant AI
Individuals and actors.
Communities

AI Deployment,
Incidents and errors are
Operation and
communicated to relevant AI
Monitoring,
actors including affected
End-Users,
Manage AI Incidents, communities. Processes for
Manage Human Factors,
4.3 Monitoring tracking, responding to, and
Domain Experts,
recovering from incidents
Affected
and errors are followed and
Individuals and
documented.
Communities

Intended purpose, potentially

beneficial uses, context-
specific laws, norms and
Socio-technical
expectations, and
systems, Societal
prospective settings in which
Values, Context of
the AI system will be
Use, Impact
deployed are understood and
Assessment, TEVV,
documented. Considerations
Trustworthy
include: specific set or types
Characteristics,
of users along with their
Validity and
Map Map 1.1 expectations; potential
Reliability, Safety,
positive and negative impacts
Secure and Resilient,
of system uses to individuals,
Accountability and
communities, organizations,
Transparency,
society, and the planet;
Explainability and
assumptions and related
Interpretability,
limitations about AI system
Privacy, Fairness and
purposes; uses and risks
Bias
across the development or
product AI lifecycle; TEVV
and system metrics.

Map Map 1.2 Diversity, Inter-disciplinary AI actors,

Interdisciplinarity, competencies, skills and
Socio-technical capacities for establishing
systems context reflect demographic
diversity and broad domain
and user experience
expertise, and their
about:blank 6/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

participation is documented.
Opportunities for
interdisciplinary collaboration
are prioritized.

The organization’s mission

Socio-technical
and relevant goals for the AI
Map Map 1.3 systems, Societal
technology are understood
Values
and documented.

The business value or context

of business use has been
Map Map 1.4 Context of Use clearly defined or – in the
case of assessing existing AI
systems – re-evaluated.

Organizational risk
Map Map 1.5 Risk Tolerance tolerances are determined
and documented.

System requirements (e.g.,

“the system shall respect the
Socio-technical privacy of its users”) are
systems, Impact elicited from and understood
Map Map 1.6
Assessment, by relevant AI actors. Design
Documentation decisions take socio-
technical implications into
account to address AI risks.

The specific task, and

methods used to implement
Socio-technical the task, that the AI system
Map Map 2.1
systems will support is defined (e.g.,
classifiers, generative
models, recommenders).

Information about the AI

system’s knowledge limits
and how system output may
be utilized and overseen by
Limitations, Human
humans is documented.
oversight, Impact
Map Map 2.2 Documentation provides
Assessment,
sufficient information to
Documentation
assist relevant AI actors
when making informed
decisions and taking
subsequent actions.

Map Map 2.3 AI Development, TEVV, Data, Impact Scientific integrity and TEVV
TEVV, Domain Assessment, considerations are identified
about:blank 7/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

Experts Limitations and documented, including
those related to experimental
design, data collection and
selection (e.g., availability,
representativeness,
suitability), system
trustworthiness, and
construct validation.

AI Development, Potential benefits of intended

Socio-technical
AI Deployment, AI system functionality and
Map Map 3.1 systems,
AI Impact performance are examined
Documentation
Assessment and documented.

Impact Assessment,
Trustworthy
Potential costs, including
Characteristics,
AI Design, AI non-monetary costs, which
Validity and
Development, result from expected or
Reliability, Safety,
Operation and realized AI errors or system
Secure and Resilient,
Map Map 3.2 Monitoring, AI functionality and
Accountability and
Design, AI trustworthiness - as
Transparency,
Impact connected to organizational
Explainability and
Assessment risk tolerance - are examined
Interpretability,
and documented.
Privacy, Fairness and
Bias

Targeted application scope is

specified and documented
AI Design, AI
Context of Use, based on the system’s
Map Map 3.3 Development,
Documentation capability, established
Human Factors
context, and AI system
categorization.

Processes for operator and

AI Design, AI
practitioner proficiency with
Development,
AI system performance and
Human Factors,
trustworthiness – and
Map Map 3.4 End-Users, Human-AI teaming
relevant technical standards
Domain Experts,
and certifications – are
Operation and
defined, assessed and
Monitoring
documented.

Human Factors, Processes for human

End-Users, oversight are defined,
Domain Experts, assessed, and documented in
Map Map 3.5 Human oversight
Operation and accordance with
Monitoring, AI organizational policies from
Design GOVERN function.

about:blank 8/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

Approaches for mapping AI

Third-party technology and legal risks of
Legal and Regulatory,
entities, its components – including
Third-party, Pre-
Procurement, the use of third-party data or
trained models,
Map Map 4.1 Operation and software – are in place,
Supply Chain, Risk
Monitoring, followed, and documented, as
Tolerance, Risky
Governance and are risks of infringement of a
Emergent Behavior
Oversight third-party’s intellectual
property or other rights.

AI Deployment, Internal risk controls for

TEVV, Operation components of the AI system
Third-party, Pre-
Map Map 4.2 and Monitoring, including third-party AI
trained models
Third-party technologies are identified
entities and documented.

Likelihood and magnitude of

AI Design, AI
each identified impact (both
Development, AI
potentially beneficial and
Deployment, AI
harmful) based on expected
Impact
use, past uses of AI systems
Assessment,
Participation, Impact in similar contexts, public
Map Map 5.1 Operation and
Assessment incident reports, feedback
Monitoring,
from those external to the
Affected
team that developed or
Individuals and
deployed the AI system, or
Communities,
other data are identified and
End-Users
documented.

AI Design,
Human Factors,
AI Deployment, Practices and personnel for
AI Impact supporting regular
Assessment, engagement with relevant AI
Operation and Participation, Impact actors and integrating
Map Map 5.2
Monitoring, Assessment feedback about positive,
Domain Experts, negative, and unanticipated
Affected impacts are in place and
Individuals and documented.
Communities,
End-Users

Measure Measure AI Development, Trustworthy Approaches and metrics for

1.1 TEVV, Domain Characteristics, Risk measurement of AI risks
Experts Assessment, Risky enumerated during the Map
Emergent Behavior, function are selected for
TEVV, Validity and implementation starting with
Reliability, Safety, the most significant AI risks.
Secure and Resilient, The risks or trustworthiness
about:blank 9/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

Accountability and characteristics that will not –
Transparency, or cannot – be measured are
Explainability and properly documented.
Interpretability,
Privacy, Fairness and
Bias

TEVV, AI Impact Appropriateness of AI

Assessment, AI metrics and effectiveness of
Development, AI existing controls is regularly
Measure Impact Assessment,
Measure Deployment, assessed and updated
1.2 TEVV, Context of Use
Affected including reports of errors
Individuals and and impacts on affected
Communities communities.

Internal experts who did not

serve as front-line developers
TEVV, AI Impact
for the system and/or
Assessment, AI
independent assessors are
Development, AI
involved in regular
Deployment,
assessments and updates.
Affected Participation, Impact
Measure Domain experts, users, AI
Measure Individuals and Assessment, Context
1.3 actors external to the team
Communities, of Use
that developed or deployed
Domain Experts,
the AI system, and affected
End-Users,
communities are consulted in
Operation and
support of assessments as
Monitoring
necessary per organizational
risk tolerance.

Test sets, metrics, and details

about the tools used during
Measure TEVV, Documentation,
Measure TEVV test, evaluation, validation,
2.1 Validity and Reliability
and verification (TEVV) are
documented.

Evaluations involving human

subjects meet applicable
TEVV, Human
Measure Data, Human Subjects requirements (including
Measure Factors, AI
2.2 Protection human subject protection)
Development
and are representative of the
relevant population.

Measure Measure TEVV, AI TEVV, Impact AI system performance or

2.3 Deployment Assessment assurance criteria are
measured qualitatively or
quantitatively and
demonstrated for conditions
similar to deployment

about:blank 10/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

setting(s). Measures are
documented.

The functionality and

behavior of the AI system and
Measure AI Deployment, TEVV, Monitoring, its components – as identified
Measure
2.4 TEVV Drift in the MAP function – are
monitored when in
production.

The AI system to be deployed

is demonstrated to be valid
TEVV, Validity and
and reliable. Limitations of
Measure TEVV, Domain Reliability,
Measure the generalizability beyond
2.5 Experts Trustworthy
the conditions under which
Characteristics, Data
the technology was
developed are documented.

AI system is evaluated
regularly for safety risks – as
identified in the MAP
function. The AI system to be
deployed is demonstrated to
TEVV, Domain
be safe, its residual negative
Experts,
TEVV, Safety, risk does not exceed the risk
Operation and
Measure Trustworthy tolerance, and can fail safely,
Measure Monitoring, AI
2.6 Characteristics, particularly if made to
Impact
Context of Use operate beyond its
Assessment, AI
knowledge limits. Safety
Deployment
metrics implicate system
reliability and robustness,
real-time monitoring, and
response times for AI system
failures.

TEVV, Domain
Experts, TEVV, Secure and
AI system security and
Operation and Resilient, Trustworthy
Measure resilience – as identified in the
Measure Monitoring, AI Characteristics,
2.7 MAP function – are evaluated
Impact Adversarial, Risky
and documented.
Assessment, AI Emergent Behavior
Deployment

about:blank 11/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

TEVV, Domain
Experts, Risks associated with
TEVV, Accountability
Operation and transparency and
Measure and Transparency,
Measure Monitoring, AI accountability – as identified
2.8 Trustworthy
Impact in the MAP function – are
Characteristics
Assessment, AI examined and documented.
Deployment

TEVV, Domain The AI model is explained,

Experts, validated, and documented,
Operation and TEVV, Explainability and AI system output is
Measure Monitoring, AI and Interpretability, interpreted within its context
Measure
2.9 Impact Trustworthy – as identified in the MAP
Assessment, AI Characteristics function – and to inform
Deployment, responsible use and
End-Users governance.

TEVV, Domain
Experts,
Operation and Privacy risk of the AI system –
TEVV, Privacy,
Measure Monitoring, AI as identified in the MAP
Measure Trustworthy
2.10 Impact function – is examined and
Characteristics
Assessment, AI documented.
Deployment,
End-Users

TEVV, Domain
Experts,
Operation and
Monitoring, AI
Fairness and bias – as
Impact TEVV, Fairness and
Measure identified in the MAP
Measure Assessment, AI Bias, Trustworthy
2.11 function – are evaluated and
Deployment, Characteristics
results are documented.
End-Users,
Affected
Individuals and
Communities

TEVV, Domain
Environmental impact and
Experts,
sustainability of AI model
Operation and
Measure TEVV, Environmental training and management
Measure Monitoring, AI
2.12 Impact activities – as identified in the
Impact
MAP function – are assessed
Assessment, AI
and documented.
Deployment

Measure Measure TEVV, AI TEVV, Effectiveness Effectiveness of the

2.13 Deployment, employed TEVV metrics and
processes in the MEASURE
about:blank 12/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

Operation and function are evaluated and
Monitoring documented.

Approaches, personnel, and

documentation are in place to
TEVV, AI Impact regularly identify and track
TEVV, Monitoring,
Measure Assessment, existing, unanticipated, and
Measure Continual
3.1 Operation and emergent AI risks based on
Improvement
Monitoring factors such as intended and
actual performance in
deployed contexts.

Risk tracking approaches are

TEVV, Domain
considered for settings where
Experts, AI
AI risks are difficult to assess
Measure Impact Monitoring, Continual
Measure using currently available
3.2 Assessment, Improvement
measurement techniques or
Operation and
where metrics are not yet
Monitoring
available.

TEVV, AI
Feedback processes for end
Deployment,
users and impacted
Operation and
Participation, communities to report
Measure Monitoring,
Measure Contestability, TEVV, problems and appeal system
3.3 End-Users,
Impact Assessment outcomes are established
Affected
and integrated into AI system
Individuals and
evaluation metrics.
Communities

TEVV, AI Measurement approaches for

Deployment, identifying AI risks are
Operation and connected to deployment
Measure Monitoring, TEVV, Participation, context(s) and informed
Measure
4.1 End-Users, Context of Use through consultation with
Affected domain experts and other end
Individuals and users. Approaches are
Communities documented.

about:blank 13/14
2/25/25, 7:21 PM Playbook

Type Title AI Actors Topics Description

TEVV, Participation, Measurement results

Trustworthy regarding AI system
Characteristics, trustworthiness in
TEVV, AI Validity and deployment context(s) and
Deployment, Reliability, Safety, across AI lifecycle are
Measure Domain Experts, Secure and Resilient, informed by input from
Measure
4.2 Operation and Accountability and domain experts and other
Monitoring, Transparency, relevant AI actors to validate
End-Users Explainability and whether the system is
Interpretability, performing consistently as
Privacy, Fairness and intended. Results are
Bias documented.

TEVV, Participation,
Trustworthy Measurable performance
TEVV, AI Characteristics, improvements or declines
Deployment, Validity and based on consultations with
Operation and Reliability, Safety, relevant AI actors including
Measure Monitoring, Secure and Resilient, affected communities, and
Measure
4.3 End-Users, Accountability and field data about context-
Affected Transparency, relevant risks and
Individuals and Explainability and trustworthiness
Communities Interpretability, characteristics, are identified
Privacy, Fairness and and documented.
Bias

about:blank 14/14