{ "schema": "yarn-pack/2", "id": "framework-nist", "version": "0.1.0", "name": "NIST AI RMF (framework self-assessment)", "engagement": "AISG — framework self-assessment: workshop prep, workshop, or diagnostic strand", "intro": "A guided conversation against NIST AI RMF, not a form. Answer in your own words and name the document or record that shows it if you can. About 22 minutes. Assesses how far an organisation’s AI risk management delivers the outcomes of the NIST AI RMF 1.0 Core (Govern, Map, Measure, Manage), its seven trustworthiness characteristics and the GenAI Profile. Certifies nothing.", "tone_default": "professional", "prefill_fields": { "department": [ "Executive", "Finance", "Operations", "Customer / Sales", "Technology / IT", "Data & Analytics", "People & Culture", "Risk & Compliance", "Marketing", "Product" ] }, "scales": [ { "id": "maturity5", "name": "Maturity (distilled model)", "levels": [ { "value": 1, "label": "Does not exist", "gloss": "No capability. Absent, or purely ad hoc / accidental." }, { "value": 2, "label": "Partially exists", "gloss": "Emerging and inconsistent. Pockets of activity, not joined up." }, { "value": 3, "label": "Fully exists", "gloss": "Defined, documented and operating across the organisation." }, { "value": 4, "label": "Fully exists & optimised", "gloss": "Measured, refined and improving against targets." }, { "value": 5, "label": "Fully exists & adaptive", "gloss": "Continuously self-adjusting; a source of advantage." } ], "signals": { "1": [ "no ", "not ", "none", "never", "don't", "do not", "nothing", "absent", "unaware", "haven't", "ad hoc", "ad-hoc", "nonexistent", "no idea", "not really" ], "2": [ "some ", "starting", "beginning", "emerging", "pilot", "trial", "informal", "inconsistent", "pockets", "a bit", "occasionally", "early", "experiment", "trying", "patchy" ], "3": [ "documented", "defined", "standard", "standardised", "established", "policy", "framework", "process", "consistent", "across the", "in place", "formal", "governed", "rolled out" ], "4": [ "measured", "metrics", "optimis", "improving", "kpi", "monitored", "reviewed", "refined", "benchmarked", "targets", "tracked", "mature", "regularly review" ], "5": [ "continuous", "adaptive", "self-", "automated end", "best in class", "best-in-class", "competitive advantage", "industry leading", "always", "real-time monitoring", "feedback loop" ] } } ], "categories": [ { "id": "govern", "name": "Govern: culture and accountability", "order": 1, "target_default": 3 }, { "id": "map", "name": "Map: context and risk identification", "order": 2, "target_default": 3 }, { "id": "measure", "name": "Measure: analyse, assess, track", "order": 3, "target_default": 3 }, { "id": "manage", "name": "Manage: prioritise, treat, respond", "order": 4, "target_default": 3 }, { "id": "trust", "name": "Trustworthy AI and the GenAI Profile", "order": 5, "target_default": 3 } ], "audiences": [ { "id": "lead", "name": "Governance / risk lead", "desc": "Owns the policy, the register or the risk framework", "deep_dive_sections": [] }, { "id": "owner", "name": "System or use-case owner", "desc": "Runs an AI system or use case day to day", "deep_dive_sections": [] }, { "id": "exec", "name": "Executive / sponsor", "desc": "Accountable for the outcome, not the mechanics", "deep_dive_sections": [] } ], "sections": [ { "id": "core", "title": "NIST RMF elements", "blurb": "Everyone answers these. The executive accountable for AI risk, the risk register owner, whoever leads generative AI use, procurement or vendor management, and a board member if the characteristics are to be board vocabulary.", "optional": false, "questions": [ { "id": "nist-gv-1", "type": "scored_text", "category": "govern", "name": "Policies, processes and accountability", "text": "If an AI system in your organisation caused harm tomorrow, who would be held to account, and what written rule would they point to for how it should have been handled?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No written AI risk policy or process; no one is named as accountable for AI risk, and AI decisions are made case by case with no record.", "3": "An approved AI risk policy and lifecycle processes exist, a named executive is accountable for AI risk, and decision rights are documented and followed.", "5": "Policy and accountabilities are reviewed on a set cycle and revised when incidents, profiles or new use types show a gap, without waiting for an audit to prompt it." }, "help": "Evidence that would show it: Approved AI policy with owner and review date; Documented decision rights across the AI lifecycle; Steering group charter and minutes; Named accountable executive for AI risk.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-gv-2", "type": "scored_text", "category": "govern", "name": "Workforce, culture and engagement", "text": "When you last introduced an AI tool that changed how people work, who was consulted beforehand, and what did the people using it know about its risks?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No AI training or awareness for staff; no consultation with affected staff or external parties before AI is deployed.", "3": "Role-based AI training runs on a schedule with completion tracked, and affected staff and external parties are consulted before deployments that affect them.", "5": "Training content and consultation practice update as new use types and risks appear, driven by feedback from staff and affected parties, not by a periodic refresh." }, "help": "Evidence that would show it: AI literacy programme with completion records; Role-based training matrix; Workforce consultation records for AI deployments.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-gv-3", "type": "scored_text", "category": "govern", "name": "Third-party AI risk", "text": "Which of your suppliers use AI to deliver what they sell you, and how would you find out if one of them changed the model behind it?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No record of which vendors supply or embed AI; contracts are silent on AI; nobody owns third-party AI risk.", "3": "A live register of vendor and embedded AI exists with a risk rating per entry, standard AI clauses sit in contracts, and a named owner reviews it on a set cadence.", "5": "The register and contract clauses adjust as vendors change models or capabilities, with material changes triggering a re-assessment before the change takes effect." }, "help": "Evidence that would show it: Vendor and third-party AI register; Standard AI clauses in vendor contracts; Boundary declaration per third-party data relationship.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-mp-1", "type": "scored_text", "category": "map", "name": "Context and categorisation", "text": "How many AI systems does your organisation run today, and for the one you are least sure about, what is it for and who uses it?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No inventory of AI systems; purpose and setting are undocumented; systems are not classified by risk and shadow AI is unknown.", "3": "A single use case register records purpose, owner, data and setting for every AI system, and each is tiered by a documented risk classification method.", "5": "The register and tiering are kept current by intake and shadow AI discovery, and tier criteria are revised as new capabilities and contexts emerge." }, "help": "Evidence that would show it: Master use case register; Intake form capturing purpose, owner, data and risk indicators; Use case risk classification methodology; Shadow AI register.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-mp-2", "type": "scored_text", "category": "map", "name": "Capabilities, impacts and affected parties", "text": "Pick one of your AI systems: who could it affect badly if it got things wrong, and where is that written down?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "Impacts on individuals and society are not considered; system capabilities and limitations are undocumented; no one asks who could be harmed.", "3": "Each AI system has documented capabilities, limitations and an impact assessment naming affected parties, with human oversight decisions recorded per use case.", "5": "Impact assessments are revisited when a system, its data or its context changes, and affected parties’ feedback changes the mapped risks." }, "help": "Evidence that would show it: Model card or equivalent with intended use and limitations; Human oversight model per use case; AI risk taxonomy covering human impact.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-ms-1", "type": "scored_text", "category": "measure", "name": "Methods and trustworthiness metrics", "text": "How do you know an AI system is working as intended? What do you measure, how often, and who sees the number?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No method or metric is defined for assessing AI risk; trustworthiness is asserted, not tested.", "3": "Documented methods and metrics exist per trustworthiness characteristic, are applied before deployment and on a schedule, with results recorded against each control.", "5": "Methods and metrics are re-selected as the system, its use and the state of the art change, and new evaluation techniques are adopted without a mandate to do so." }, "help": "Evidence that would show it: Controls library with verification mechanisms; Assurance schedule of model evaluations; Verification log of control test results.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-ms-2", "type": "scored_text", "category": "measure", "name": "Ongoing monitoring and feedback", "text": "If one of your AI systems had slowly started giving worse answers over the last three months, what would have told you, and who would have noticed first?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No monitoring after go-live; no channel for users or affected parties to report problems; drift and degradation go unnoticed.", "3": "Production systems are monitored against defined metrics, feedback channels exist and are reviewed, and vendor material changes trigger re-assessment.", "5": "Monitoring thresholds and feedback loops adjust as behaviour shifts, and outcomes feed back into the mapped risks and the metrics chosen." }, "help": "Evidence that would show it: Verification log and monitoring outputs; Material change notification process; Benefits realisation tracking per use case.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-mg-1", "type": "scored_text", "category": "manage", "name": "Prioritisation and risk treatment", "text": "Of the AI risks you have identified, which three worry you most right now, and what has been done about each since you named them?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "Risks are listed but not ranked or treated; no risk appetite exists to decide against; treatments are undocumented.", "3": "A live AI risk register ranks risks, records a treatment and residual rating per entry, and treatments are chosen against an endorsed risk appetite statement.", "5": "Priorities and treatments shift as measurement results and incidents arrive, with the register updated within the operating cycle rather than at annual review." }, "help": "Evidence that would show it: AI risk appetite statement; Risk register entries for AI initiatives; Controls library mapping risks to controls.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-mg-2", "type": "scored_text", "category": "manage", "name": "Third-party response and incident recovery", "text": "Tell me about the last time an AI system did something it should not have. What happened next, and what changed afterwards?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No AI incident definition or response procedure; incidents are handled ad hoc; no process to act when a vendor’s AI fails or changes.", "3": "An AI incident procedure covers definition, classification, escalation, remediation and learning; an incident register is kept; vendor changes have a response path.", "5": "Incident lessons change controls, monitoring and mapped risks as a matter of course, and response plans are rehearsed and revised as new failure modes appear." }, "help": "Evidence that would show it: AI incident response SOP; AI incident register with lessons learned; Decommissioning checklist; Material change notification process.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-tc-1", "type": "scored_text", "category": "trust", "name": "Seven characteristics of trustworthy AI", "text": "If a board member asked you what good AI means for this organisation, what words would you use, and are they the same words the people building it would use?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "No shared definition of what good AI means; trustworthiness is discussed in vague terms or not at all.", "3": "The seven characteristics are adopted in policy and used to structure risk taxonomy, position statements and disclosure, with tensions between them decided deliberately.", "5": "The characteristics are the working vocabulary from board to build team, and the balance struck between them is revisited per use case as evidence arrives." }, "help": "Evidence that would show it: AI position statement referencing the characteristics; AI risk taxonomy aligned to the characteristics; Disclosure and transparency mechanism.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false }, { "id": "nist-tc-2", "type": "scored_text", "category": "trust", "name": "Generative AI Profile (NIST AI 600-1)", "text": "Where is generative AI used in your organisation today, and for one of those uses, what stops it making things up, leaking data or producing content you would not want your name on?", "scale": "maturity5", "scored": true, "star": false, "rubric": { "1": "Generative AI is used without any GenAI-specific risk view; acceptable use is silent on it and retrieval content is uncurated.", "3": "The 12 categories are worked through per GenAI use case, with acceptable use rules, approved tooling and corpus controls in place and evidence of each category addressed.", "5": "GenAI risk treatment updates as models, profiles and usage change, and new categories or NIST addenda are folded into the checklist as they land." }, "help": "Evidence that would show it: Acceptable use guidelines covering generative AI; Approved tooling and model list; Corpus register with ownership and refresh cadence; Content sensitivity tagging.", "adaptive": { "allow_probe": true, "allow_skip": false, "max_probes": 1 }, "ai_drafted": false } ] } ], "grids": {}, "outputs": [ "Level per element and per category, gated", "Contested-element view (spread of 2 or more)", "Coverage of evidence: confirmed, stated, inferred", "Where to start, foundations first", "Printable client report" ], "report_defaults": [ "rpt-maturity-standard" ], "playbook": { "sequence": [ "govern", "map", "measure", "manage", "trust" ], "sequence_note": "Govern wraps the other three. Settle accountability first, then walk Map, Measure, Manage; use the seven characteristics as board vocabulary throughout and the GenAI Profile as the workshop checklist.", "actions": { "govern": { "to_3": [ "Name an accountable executive, approve an AI risk policy and document decision rights", "Register vendor and embedded AI and put AI clauses in contracts", "Run role-based AI training and consult affected staff before deployment" ], "to_5": [ "Use the NIST crosswalk to place Govern inside 42001 Clauses 5 and 6 so a verifiable system carries it", "Review policy and accountabilities whenever incidents or new profiles expose a gap" ] }, "map": { "to_3": [ "Stand up one use case register with purpose, owner, data and risk tier per system", "Document capabilities, limitations and affected parties per system" ], "to_5": [ "Feed intake and shadow AI discovery into the register continuously; revisit impacts when context changes" ] }, "measure": { "to_3": [ "Pick a method and metric per trustworthiness characteristic from the AI RMF Playbook; record results", "Monitor production systems and open a feedback channel for users and affected parties" ], "to_5": [ "Re-select metrics as systems and the state of the art move; let outcomes revise the mapped risks" ] }, "manage": { "to_3": [ "Endorse a risk appetite, rank the register and record a treatment per risk", "Write and test an AI incident response procedure with a register and a vendor change path" ], "to_5": [ "Let incident lessons change controls and monitoring as routine; rehearse response plans" ] }, "trust": { "to_3": [ "Adopt the seven characteristics in policy and taxonomy; run the GenAI Profile’s 12 categories per GenAI use case" ], "to_5": [ "Track NIST profiles and addenda (AI Agent Interoperability Profile due Q4 2026) and fold them in as they land" ] } } } }