Risk Assessment

NIST AI RMF: MAP-1–5, MEASURE-1–3
EU AI Act: Art. 9 (Risk management system), Annex III classification
ISO 42001: A.4.1–A.4.4 (AI risk assessment, treatment, acceptance)
NIST AI 600-1: §2.1–§2.12 risk area mapping

1. Context Establishment (MAP-1)

Element	Description	Artefact
AI System Scope	Name, version, purpose, intended use, deployment context	System card
Stakeholders	Users, subjects, operators, regulators, auditors	Stakeholder register
Regulatory Context	Applicable laws (EU AI Act, GDPR, sectoral, US state)	Regulatory applicability matrix
Risk Appetite	Organisational tolerance per category (safety, privacy, IP, reputation)	Risk appetite statement

2. Risk Identification (MAP-2)

Methodologies

Method	When to Use	Output
Structured brainstorming	Early design; cross-functional team	Risk register v1
Threat modelling (ATLAS/OWASP)	Architecture review	Threat-modelled attack trees
Scenario analysis	High-impact, low-likelihood events	Scenario library
Historical incident review	Post-deployment; continuous	Incident-derived risks
Red-team / adversarial testing	Pre-deployment; quarterly	Findings with exploit evidence

Risk Register Template

ID	Asset	Threat (ATLAS)	Vulnerability	Likelihood	Impact	Inherent Risk	Controls	Residual Risk	Owner
R-001	LLM API	T0012.003 (Prompt injection)	No input classifier	High	High	Critical	Classifier, hierarchy, encoding	Medium	ML Platform
R-002	Training data	T0012.002 (Poisoning)	Unsigned public corpus	Medium	High	High	Signed datasets, robust agg	Low	Data Eng

3. Risk Analysis & Evaluation (MAP-3, MEASURE-1)

Likelihood Scale

Level	Description	Quantitative Anchor
Very High	Expected routinely	>1×/week
High	Likely monthly	1–4×/month
Medium	Possible quarterly	1–4×/quarter
Low	Unlikely annually	<1×/year
Very Low	Exceptional	<1×/5 years

Impact Scale (per EU AI Act / ISO 42001)

Level	Safety	Fundamental Rights	Financial	Reputation	Compliance
Critical	Fatality	Severe rights violation	>10% revenue	Global scandal	Criminal liability
High	Serious injury	Major rights impact	1–10% revenue	National media	Major fines (Art. 99)
Medium	Injury	Moderate rights impact	0.1–1% revenue	Industry attention	Regulatory action
Low	Minor harm	Limited rights impact	<0.1% revenue	Internal	Minor finding
Negligible	None	None	Negligible	None	None

Risk Matrix

	Negligible	Low	Medium	High	Critical
Very High	Low	Medium	High	Critical	Critical
High	Low	Medium	High	Critical	Critical
Medium	Low	Low	Medium	High	Critical
Low	Low	Low	Low	Medium	High
Very Low	Low	Low	Low	Low	Medium

4. Risk Treatment (MAP-4, MANAGE-1)

Option	When	Example
Avoid	Risk > appetite; no viable controls	Don't deploy biometric ID in public spaces (Art. 5)
Mitigate	Controls reduce to acceptable	Input classifier + hierarchy for prompt injection
Transfer	Insurance; contractual	Cyber insurance for model extraction liability
Accept	Residual ≤ appetite; documented	Low-likelihood model DoS with rate limiting

Treatment Plan Columns: Risk ID, Option, Control Description, Implementation Date, Owner, Verification Method, Residual Risk, Review Date

5. EU AI Act Alignment

AI Act Requirement	RMF Mapping	Evidence
Art. 9 Risk management system	MAP-1–4, MANAGE-1–4	Documented risk register, treatment plan, review logs
Art. 10 Data governance	MAP-2.3, MEASURE-2	Data sheets, provenance, bias assessment
Art. 11 Technical documentation	GOVERN-1.3, MAP-2.1	System card, model card, data sheet
Art. 14 Human oversight	MANAGE-4.2	Oversight procedures, escalation paths
Art. 15 Accuracy, robustness, cybersecurity	MEASURE-2, MANAGE-2	Adversarial test results, drift metrics
Art. 17 Quality management system	GOVERN-1–6, MANAGE-4	QMS certification (ISO 9001/42001)

6. Review & Monitoring (MAP-5, MEASURE-3, MANAGE-4)

Cadence	Activity	Participants
Continuous	Drift monitoring, anomaly alerts	ML Platform, Security
Weekly	Incident review, new threat intel	SecOps, ML Eng
Monthly	Risk register update, control effectiveness	Risk Owner, CISO
Quarterly	Red-team, adversarial testing, residual risk review	Red Team, Risk, Legal
Annually	Full risk reassessment, regulatory scan	Executive, Board

Citation Register

Claim	Instrument	Provision	URL
MAP function	NIST AI 100-1	MAP-1–5	https://doi.org/10.6028/NIST.AI.100-1
MEASURE function	NIST AI 100-1	MEASURE-1–3	https://doi.org/10.6028/NIST.AI.100-1
Risk management	EU AI Act	Art. 9	https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Data governance	EU AI Act	Art. 10	https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Technical docs	EU AI Act	Art. 11	https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Human oversight	EU AI Act	Art. 14	https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Robustness	EU AI Act	Art. 15	https://eur-lex.europa.eu/eli/reg/2024/1689/oj
QMS	EU AI Act	Art. 17	https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Risk assessment	ISO 42001	A.4.1–A.4.4	https://www.iso.org/standard/81230.html
GAI risk areas	NIST AI 600-1	§2.1–§2.12	https://doi.org/10.6028/NIST.AI.600-1