Regulatory Mapping Tool
When to activate
When identifying applicable regulations for AI systems, designing regulatory compliance controls, mapping specific regulatory requirements to operational procedures, assessing compliance gaps, or preparing for regulatory audits.
When NOT to use
For legal advice or formal regulatory interpretation. This skill is for operational mapping: identifying which rules apply and designing compliance controls. For regulatory questions, consult qualified legal counsel.
Instructions
Regulatory Mapping Process
Define System Scope
- Model name and business use case
- Geographic scope: Where is model deployed? (US, EU, China, Canada, etc.)
- Industry: What sector? (Finance, Healthcare, Employment, Consumer, etc.)
- User/affected population: Employees, customers, vulnerable groups?
- Decision type: Consequential (loan, hiring, medical) or advisory (recommendation)?
Identify Applicable Regulations
Use this framework to identify relevant regulations:
Jurisdiction-Based:
- EU: GDPR, AI Act, sector-specific (ePrivacy, financial regulations, healthcare)
- US: FTC Act, FCRA, HIPAA, EEOC employment rules, NIST AI RMF guidance, state laws (CCPA, etc.)
- Global: Consider where data flows, where model is deployed
Industry-Specific:
- Finance: SEC disclosure rules, FINRA, banking regulations, Fair Lending laws (FCRA, ECOA)
- Healthcare: HIPAA, FDA (if medical device), state licensing boards
- Employment: Title VII, EEOC guidance, state hiring laws, anti-discrimination rules
- Consumer Protection: FTC Act, CCPA, state consumer protection laws
- Automated Decision-Making: AI Act (EU), state algorithmic transparency laws
Risk-Based:
- High-stakes decisions: Additional scrutiny (hiring, credit, criminal justice)
- Vulnerable populations: Children, elderly, low-income require heightened protection
- PII/sensitive data: GDPR, CCPA, HIPAA, state privacy laws
Decision type-based:
- Consequential (deny/approve decisions): Fair lending, employment discrimination, due process rules apply
- Ranking/scoring: Transparency and explainability requirements
- Advisory: Fewer compliance requirements but still governance needed
Map Regulatory Requirements to Specific Controls
For each applicable regulation, identify specific requirements:
Example: GDPR Requirements
| Article |
Requirement |
Operational Control |
| Article 5(1)(a) |
Lawful processing |
Document lawful basis (consent, legitimate interest, contract, etc.); evidence of compliance |
| Article 5(1)(b) |
Specified purpose |
Define AI system purpose in documentation; do not use for unrelated purposes |
| Article 5(1)(c) |
Data minimization |
Use only features necessary for model purpose; justify each feature |
| Article 5(1)(f) |
Security |
Encrypt PII; access controls; breach notification plan |
| Article 13/14 |
Transparency |
Privacy notice to users explaining model and data use |
| Article 22 |
Right to explanation |
Provide explanation of model decisions when consequential |
| Article 35 |
DPIA |
Conduct Data Protection Impact Assessment for high-risk processing |
| Article 37 |
DPO |
Appoint Data Protection Officer if processing requires |
Example: AI Act Requirements (EU) - High-Risk Systems
| Requirement |
Operational Control |
| Risk classification |
Classify model as prohibited, high-risk, or low-risk; document assessment |
| Technical documentation |
Maintain complete documentation of training, testing, deployment |
| Conformity assessment |
Third-party assessment required for high-risk systems |
| Quality management |
Implement QMS for AI development and monitoring |
| Bias and fairness monitoring |
Establish metrics and monitoring for discrimination |
| Transparency and user info |
Disclose that AI is making decisions; explain how it works |
| Human oversight |
Establish mechanisms for human review of decisions |
| Record-keeping |
Maintain audit logs and decision records for 3 years |
Example: FCRA Requirements (US Fair Lending)
| Requirement |
Operational Control |
| Permissible purpose |
Ensure model used only for credit decisions (not other purposes) |
| Adverse action notice |
When credit is denied, provide notice explaining why; right to dispute |
| Dispute procedures |
Establish process for consumers to dispute model decisions and correct data |
| Accuracy |
Validate training data accuracy; monitor model for errors |
| Non-discrimination |
Ensure model does not discriminate based on protected class (race, gender, age, etc.) |
Example: EEOC Guidance on AI in Hiring (US)
| Requirement |
Operational Control |
| Disparate impact analysis |
Monitor hiring outcomes by protected class (race, gender, age) |
| Transparency |
Job applicants informed that AI is used in hiring; high-performing candidates subject to human review |
| Validation |
Ensure model predicts job performance; establish baseline (human hiring performance) and compare |
| Record-keeping |
Maintain records of hiring decisions, model inputs, outcomes by protected class for 1+ years |
| Reasonable accommodation |
Process for applicants to request accommodations if AI excludes them |
Assess Compliance Gaps
For each requirement, evaluate:
- Is control currently implemented? (Yes/No/Partial)
- What evidence exists? (Policy, procedure, audit trail, logs)
- What is the gap? (Missing, undocumented, untested)
- What is the risk if not addressed? (Regulatory enforcement, litigation, reputational)
Gap Assessment Template:
| Regulation |
Requirement |
Current Status |
Evidence |
Gap |
Risk Level |
Remediation |
| GDPR |
Lawful basis documented |
Partial |
Consent form signed; not documented in records |
Consent not logged/tracked |
Medium |
Create consent tracking database |
| GDPR |
DPIA completed |
No |
N/A |
No DPIA started |
High |
Conduct DPIA by [Date] |
| AI Act |
Risk classification |
Yes |
Risk assessment completed |
N/A |
Low |
Update quarterly |
| FCRA |
Adverse action notice |
Yes |
Process documented |
Unclear if all denials get notice |
Medium |
Audit process; test with sample |
| EEOC |
Disparate impact analysis |
Partial |
Quarterly monitoring report |
Limited to hiring outcomes; not updated since Q1 |
High |
Establish monthly monitoring dashboard |
Design Compliance Roadmap
Prioritize and sequence remediation:
Priority 1 (Critical - implement before deployment):
- High-risk gaps in critical regulations
- Gaps affecting consequential decisions (loan, hiring, medical)
- Gaps in fairness/anti-discrimination controls
Priority 2 (High - implement within 90 days):
- Medium-risk gaps in core regulations
- Gaps in documentation/transparency
- Gaps in monitoring procedures
Priority 3 (Medium - implement within 6 months):
- Low-risk gaps
- Gaps in less-critical regulations
- Enhancements beyond compliance baseline
Document Regulatory Assessment
Prepare Regulatory Mapping document including:
- Applicable regulations identified
- Regulatory requirements mapped to controls
- Compliance assessment (current state)
- Gaps and risk assessment
- Remediation plan with timeline
- Sign-off by Legal and Compliance
Example
Scenario: US fintech company deploying AI model to predict loan defaults. Model used for pre-approval screening. Model affects both consumer loan applicants and business loan decisions.
# Regulatory Mapping: Loan Default Prediction Model
## 1. System Scope
**Model:** Loan Default Risk Predictor (LDP)
**Business Purpose:** Pre-approval scoring for consumer auto loans and small business lines of credit
**Geographic Scope:** All 50 US states + DC; some international customers (secondary market)
**Affected Parties:** Consumer loan applicants (primary; FCRA protected); business applicants (secondary; less regulated)
**Decision Type:** Consequential (affects loan approval/denial/interest rate)
**Data Scope:** Credit bureau data, bank account data, transaction data
---
## 2. Applicable Regulations - Identified
### Primary (Direct Impact)
**Federal Fair Lending Laws:**
- Fair Credit Reporting Act (FCRA, 15 U.S.C. § 1681)
- Equal Credit Opportunity Act (ECOA, 15 U.S.C. § 1691)
- Fair Housing Act (FHA, 42 U.S.C. § 3601) - if used for mortgage or housing-related credit
- Community Reinvestment Act (CRA) - if depository institution
**Consumer Privacy Laws:**
- Gramm-Leach-Bliley Act (GLBA) - financial privacy rules
- California Consumer Privacy Act (CCPA) - if California customers
- State privacy laws (NY BitLicense, MA requirements, etc.)
**AI/Algorithmic Transparency Laws:**
- NY Local Law 144 (algorithmic accountability; requires notice that AI used in decisions)
- State consumer protection laws (FTC Act enforced by states)
- NIST AI Risk Management Framework (guidance, not binding but increasingly expected)
**Industry Guidance:**
- FDIC/Fed/OCC Fair Lending Guidance (2023)
- EEOC AI Guidance (emerging standards for employment; not directly applicable but principles informative)
### Secondary (Indirect)
**Data Security:**
- State data breach notification laws
- PCI DSS (if credit card data handled)
- FTC Safeguards Rule (for financial institutions)
**International (if applicable):**
- GDPR (if EU customers; less likely for consumer loans but assess)
- UK AI Bill (if UK expansion planned)
---
## 3. Regulatory Requirements Mapped to Controls
### FCRA Compliance (Fair Credit Reporting Act)
| FCRA Requirement | Application to LDP Model | Operational Control Required | Current Status | Gap |
|---|---|---|---|---|
| **Permissible Purpose** | Model must be used ONLY for credit decisions (loan/credit limit/interest rate). Cannot use for marketing, employment, insurance, etc. | Document model's permitted use case; restrict model API access to loan decisions only | Yes — documented in model card | None |
| **Accuracy** | Ensure model training data is accurate; monitor for systematic errors | Validate training data sources; quarterly accuracy audits; retraining pipeline | Partial — accuracy audit not currently scheduled | Medium: Implement quarterly audit schedule |
| **Adverse Action Notice** | When loan is DENIED (or credit terms unfavorable), provide consumer with notice explaining the reason and right to dispute | When model score triggers decline, system must provide adverse action notice (not just model score) | Partial — adverse action sent but insufficient explanation of model factors | Medium: Enhance notice to explain top factors influencing score |
| **Dispute Procedures** | Consumer can dispute accuracy of data/decision; company must reinvestigate | Establish process for loan applicants to dispute model decisions; escalate to human review | No — not currently implemented | High: Critical for compliance; implement dispute intake form and process |
| **Non-Discrimination** | Model cannot discriminate based on protected class (race, color, religion, national origin, sex, marital/family status, age, receipt of public assistance) | Monitor outcomes by protected class; conduct disparate impact analysis; identify proxy variables | Partial — demographic monitoring attempted but incomplete; no disparate impact analysis | High: Implement comprehensive disparate impact testing |
| **Fair Credit Reporting Practices** | Follow FTC guidelines for consumer information use | Privacy notice (privacy policy), consent practices, data security | Yes — privacy policy and consent in place | None |
**FCRA Assessment:** Partial compliance. Critical gaps in dispute procedures and disparate impact analysis.
---
### ECOA Compliance (Equal Credit Opportunity Act)
| ECOA Requirement | Application to LDP Model | Control | Status | Gap |
|---|---|---|---|---|
| **Non-discrimination** | Cannot discriminate based on sex, marital status, age, race, color, religion, national origin, receipt of public assistance, or exercise of consumer rights | Ensure model not trained on protected class features; monitor for proxy discrimination | Partial — no protected class features used explicitly, but proxy risk assessed | Medium: Document proxy variable risk; implement monitoring |
| **Adverse Action Notice** | Similar to FCRA; consumer right to know why credit was denied | See FCRA adverse action notice requirement | Partial | Medium: (same as FCRA) |
| **Appraisal Independence** (if mortgage-related) | Independent appraisals required; model cannot override | Not applicable to auto/business loans | N/A | None |
**ECOA Assessment:** Largely aligned with FCRA compliance. Focus on disparate impact and proxy discrimination.
---
### NY Local Law 144 (Algorithmic Accountability; if customers in NY)
| Requirement | Application to LDP | Control | Status | Gap |
|---|---|---|---|---|
| **Notice** | Notify consumers that automated decision system is used in hiring (not directly applicable to credit but principle is important) | For credit decisions, disclose that AI is used; explain how it works | Partial — privacy policy mentions model but not clear or prominent | Low: Add clear disclosure to loan application and adverse action notice |
| **Annual Bias Audit** | Audit for bias and disparate impact; document findings | Conduct and document annual fairness/bias audit | No — not currently audited | High: Critical for compliance; implement annual audit by [Date] |
| **Data Minimization** | Use only necessary data; justify use of each data source | Document why each data source is used; remove unnecessary features | Partial — features justified but not systematically reviewed | Medium: Conduct feature justification review |
**LL 144 Assessment:** Significant gaps. Not yet compliant; must implement before full deployment.
---
### State Consumer Protection / GLBA (Financial Privacy)
| Requirement | Control | Status | Gap |
|---|---|---|---|
| Privacy notice (what data you collect and how you use it) | Detailed privacy policy; model use disclosed | Yes | None |
| Consent (for some data sharing) | Opt-in consent for sharing with third parties | Yes — privacy policy includes consent mechanism | None |
| Data security (reasonable security measures) | Encryption, access controls, breach notification plan | Partial — encryption in place; access controls limited | Medium: Strengthen access controls |
| Breach notification | Notify affected users if breach occurs | Process documented | None |
---
## 4. Compliance Gaps & Risk Assessment
### High-Risk Gaps (Must remediate before deployment)
**Gap 1: Disparate Impact Analysis**
- **Regulation:** FCRA, ECOA
- **Requirement:** Monitor outcomes for discrimination; ensure model does not have disparate impact
- **Current State:** No systematic monitoring
- **Risk Level:** High — regulatory enforcement risk, litigation risk, reputational risk
- **Remediation:**
1. Conduct baseline disparate impact analysis (identify protected classes; compare outcomes)
2. Establish quarterly monitoring
3. Design mitigation if disparate impact detected (e.g., fairness constraints, manual review threshold)
4. **Timeline:** By [Deployment Date - 4 weeks]; ongoing quarterly
**Gap 2: Dispute Procedures**
- **Regulation:** FCRA
- **Requirement:** Provide process for consumers to dispute decisions
- **Current State:** No documented process
- **Risk Level:** High — direct FCRA violation; regulatory enforcement likely
- **Remediation:**
1. Design dispute intake process
2. Document reinvestigation procedure
3. Train staff on dispute handling
4. Log all disputes and outcomes for audit trail
5. **Timeline:** By [Deployment Date - 2 weeks]; before going live
**Gap 3: Bias Audit Procedures**
- **Regulation:** NY LL 144, FTC guidance
- **Requirement:** Conduct annual bias audit; document findings
- **Current State:** Audit not scheduled; no procedures defined
- **Risk Level:** Medium (regulatory expectation, not yet legal requirement nationally)
- **Remediation:**
1. Define audit scope, methodology, protected classes
2. Schedule annual audit before [Date]
3. Document findings and remediation actions
4. **Timeline:** First audit by [Date + 90 days]; then annually
### Medium-Risk Gaps (Remediate within 90 days of deployment)
**Gap 4: Enhanced Adverse Action Notices**
- **Regulation:** FCRA
- **Requirement:** Notice must explain decision; currently sent but lacks detail
- **Risk Level:** Medium — currently compliant but opportunity to enhance
- **Remediation:** Update notice template to include top 3 model factors; example: "Your application was not approved primarily due to: (1) High debt-to-income ratio, (2) Limited credit history, (3) Recent delinquency. You have the right to..."
- **Timeline:** Implement by [Deployment + 30 days]
**Gap 5: Access Control Enhancement**
- **Regulation:** GLBA, GDPR (if EU customers)
- **Requirement:** Restrict data access; log all access
- **Risk Level:** Medium — security and privacy risk
- **Remediation:** Implement field-level access control; log all credit bureau data access
- **Timeline:** Implement by [Deployment + 90 days]
---
## 5. Regulatory Compliance Status Summary
| Regulation | Critical Requirement | Status | Gap Level | Remediation Timeline |
|---|---|---|---|---|
| FCRA | Adverse action notice | Partial | Medium | Deployment + 30 days |
| FCRA | Dispute procedures | No | High | Before deployment |
| FCRA | Accuracy monitoring | Partial | Medium | Q1 Year 2 |
| FCRA | Permissible purpose | Yes | None | — |
| ECOA | Non-discrimination | Partial | High | Before deployment |
| ECOA | Disparate impact analysis | No | High | Before deployment |
| NY LL 144 | Notice of AI use | Partial | Low | Deployment + 30 days |
| NY LL 144 | Annual bias audit | No | Medium | Deployment + 90 days |
| NY LL 144 | Data minimization | Partial | Low | Deployment + 60 days |
| GLBA | Privacy & consent | Yes | None | — |
| GLBA | Data security | Partial | Medium | Deployment + 90 days |
**Overall Compliance Status:** CONDITIONAL APPROVAL
- **Before Deployment:** Remediate dispute procedures (Gap 2) and disparate impact analysis (Gap 1)
- **By Deployment + 30 days:** Remediate adverse action notices (Gap 4) and AI use disclosure (LL 144)
- **By Deployment + 90 days:** Remediate bias audit (Gap 3), data minimization (Gap 5), access controls (Gap 5)
---
## 6. Regulatory Compliance Roadmap
BEFORE DEPLOYMENT (Critical)
├─ Disparate impact analysis: complete baseline assessment
├─ Dispute procedures: design and document process
├─ Staff training: ensure staff understand FCRA/ECOA requirements
└─ Legal review: final sign-off on compliance posture
DEPLOYMENT + 30 DAYS (High Priority)
├─ Adverse action notice: implement enhanced template with model factors
├─ AI use disclosure: add clear disclosure to loan app and notices
└─ Monitoring dashboard: implement to track outcomes by protected class
DEPLOYMENT + 90 DAYS (Medium Priority)
├─ Bias audit procedures: define and document annual audit process
├─ Access control: implement field-level restrictions and logging
├─ Data minimization: review features; document justification for each
└─ Incident response: finalize procedures for regulatory inquiries
ONGOING (Continuous)
├─ Quarterly: disparate impact monitoring (protected class outcomes)
├─ Quarterly: adverse action audit (sample 10 denials; verify notice provided)
├─ Annually: bias and fairness audit (full assessment)
├─ Continuously: dispute intake and reinvestigation (log all)
└─ Continuously: incident response (escalate any signs of discrimination)
---
## 7. Approvals & Sign-Off
**Regulatory Compliance Assessment completed:** [Date]
- **Legal Counsel:** _____________________ Date: _____ [Confirms regulatory assessment and remediation plan]
- **Compliance Officer:** _____________________ Date: _____ [Confirms compliance roadmap feasible and resourced]
- **Model Owner:** _____________________ Date: _____ [Confirms ability to implement technical controls]
- **Executive Sponsor:** _____________________ Date: _____ [Approves deployment contingent on Gap 1 & 2 remediation]
**Deployment Authorized:** [Date] pending completion of Critical remediation items
**Follow-up Audit Scheduled:** [Date + 120 days]
---
## Regulatory Reference
**Federal:**
- FCRA: 15 U.S.C. § 1681
- ECOA: 15 U.S.C. § 1691
- GLBA: 15 U.S.C. § 6801
- FDIC/Fed/OCC Fair Lending Guidance (2023)
**State:**
- NY LL 144: Local Law 144 (2023)
- CA CCPA: California Consumer Privacy Act (2018)
- MA AI Transparency: "An Act Relative to the Oversight of Artificial Intelligence" (2023)
**International:**
- GDPR: Regulation (EU) 2016/679
- UK AI Bill (pending)
**Guidance:**
- NIST AI Risk Management Framework (2023)
- FTC "Endorsement Guides" for AI recommendations
Regulatory Mapping Checklist
Ensure comprehensive assessment:
Regulatory Identification:
Requirements Mapping:
Gap Assessment:
Compliance Roadmap:
Documentation & Approval:
1---2name: regulatory-mapping-tool3description: Regulatory Mapping Tool4metadata:5 source: https://skillmd.com/skills/uitbreidenos/regulatory-mapping-tool6---7# Regulatory Mapping Tool89## When to activate1011When identifying applicable regulations for AI systems, designing regulatory compliance controls, mapping specific regulatory requirements to operational procedures, assessing compliance gaps, or preparing for regulatory audits.1213## When NOT to use1415For legal advice or formal regulatory interpretation. This skill is for operational mapping: identifying which rules apply and designing compliance controls. For regulatory questions, consult qualified legal counsel.1617## Instructions1819### Regulatory Mapping Process20211. **Define System Scope**22 - Model name and business use case23 - Geographic scope: Where is model deployed? (US, EU, China, Canada, etc.)24 - Industry: What sector? (Finance, Healthcare, Employment, Consumer, etc.)25 - User/affected population: Employees, customers, vulnerable groups?26 - Decision type: Consequential (loan, hiring, medical) or advisory (recommendation)?27282. **Identify Applicable Regulations**29 30 Use this framework to identify relevant regulations:31 32 **Jurisdiction-Based:**33 - EU: GDPR, AI Act, sector-specific (ePrivacy, financial regulations, healthcare)34 - US: FTC Act, FCRA, HIPAA, EEOC employment rules, NIST AI RMF guidance, state laws (CCPA, etc.)35 - Global: Consider where data flows, where model is deployed36 37 **Industry-Specific:**38 - **Finance:** SEC disclosure rules, FINRA, banking regulations, Fair Lending laws (FCRA, ECOA)39 - **Healthcare:** HIPAA, FDA (if medical device), state licensing boards40 - **Employment:** Title VII, EEOC guidance, state hiring laws, anti-discrimination rules41 - **Consumer Protection:** FTC Act, CCPA, state consumer protection laws42 - **Automated Decision-Making:** AI Act (EU), state algorithmic transparency laws43 44 **Risk-Based:**45 - **High-stakes decisions:** Additional scrutiny (hiring, credit, criminal justice)46 - **Vulnerable populations:** Children, elderly, low-income require heightened protection47 - **PII/sensitive data:** GDPR, CCPA, HIPAA, state privacy laws48 49 **Decision type-based:**50 - **Consequential (deny/approve decisions):** Fair lending, employment discrimination, due process rules apply51 - **Ranking/scoring:** Transparency and explainability requirements52 - **Advisory:** Fewer compliance requirements but still governance needed53543. **Map Regulatory Requirements to Specific Controls**55 56 For each applicable regulation, identify specific requirements:57 58 **Example: GDPR Requirements**59 60 | Article | Requirement | Operational Control |61 |---------|-------------|---|62 | Article 5(1)(a) | Lawful processing | Document lawful basis (consent, legitimate interest, contract, etc.); evidence of compliance |63 | Article 5(1)(b) | Specified purpose | Define AI system purpose in documentation; do not use for unrelated purposes |64 | Article 5(1)(c) | Data minimization | Use only features necessary for model purpose; justify each feature |65 | Article 5(1)(f) | Security | Encrypt PII; access controls; breach notification plan |66 | Article 13/14 | Transparency | Privacy notice to users explaining model and data use |67 | Article 22 | Right to explanation | Provide explanation of model decisions when consequential |68 | Article 35 | DPIA | Conduct Data Protection Impact Assessment for high-risk processing |69 | Article 37 | DPO | Appoint Data Protection Officer if processing requires |70 71 **Example: AI Act Requirements (EU) - High-Risk Systems**72 73 | Requirement | Operational Control |74 |---|---|75 | Risk classification | Classify model as prohibited, high-risk, or low-risk; document assessment |76 | Technical documentation | Maintain complete documentation of training, testing, deployment |77 | Conformity assessment | Third-party assessment required for high-risk systems |78 | Quality management | Implement QMS for AI development and monitoring |79 | Bias and fairness monitoring | Establish metrics and monitoring for discrimination |80 | Transparency and user info | Disclose that AI is making decisions; explain how it works |81 | Human oversight | Establish mechanisms for human review of decisions |82 | Record-keeping | Maintain audit logs and decision records for 3 years |83 84 **Example: FCRA Requirements (US Fair Lending)**85 86 | Requirement | Operational Control |87 |---|---|88 | Permissible purpose | Ensure model used only for credit decisions (not other purposes) |89 | Adverse action notice | When credit is denied, provide notice explaining why; right to dispute |90 | Dispute procedures | Establish process for consumers to dispute model decisions and correct data |91 | Accuracy | Validate training data accuracy; monitor model for errors |92 | Non-discrimination | Ensure model does not discriminate based on protected class (race, gender, age, etc.) |93 94 **Example: EEOC Guidance on AI in Hiring (US)**95 96 | Requirement | Operational Control |97 |---|---|98 | Disparate impact analysis | Monitor hiring outcomes by protected class (race, gender, age) |99 | Transparency | Job applicants informed that AI is used in hiring; high-performing candidates subject to human review |100 | Validation | Ensure model predicts job performance; establish baseline (human hiring performance) and compare |101 | Record-keeping | Maintain records of hiring decisions, model inputs, outcomes by protected class for 1+ years |102 | Reasonable accommodation | Process for applicants to request accommodations if AI excludes them |1031044. **Assess Compliance Gaps**105 106 For each requirement, evaluate:107 - Is control currently implemented? (Yes/No/Partial)108 - What evidence exists? (Policy, procedure, audit trail, logs)109 - What is the gap? (Missing, undocumented, untested)110 - What is the risk if not addressed? (Regulatory enforcement, litigation, reputational)111 112 **Gap Assessment Template:**113 114 | Regulation | Requirement | Current Status | Evidence | Gap | Risk Level | Remediation |115 |---|---|---|---|---|---|---|116 | GDPR | Lawful basis documented | Partial | Consent form signed; not documented in records | Consent not logged/tracked | Medium | Create consent tracking database |117 | GDPR | DPIA completed | No | N/A | No DPIA started | High | Conduct DPIA by [Date] |118 | AI Act | Risk classification | Yes | Risk assessment completed | N/A | Low | Update quarterly |119 | FCRA | Adverse action notice | Yes | Process documented | Unclear if all denials get notice | Medium | Audit process; test with sample |120 | EEOC | Disparate impact analysis | Partial | Quarterly monitoring report | Limited to hiring outcomes; not updated since Q1 | High | Establish monthly monitoring dashboard |1211225. **Design Compliance Roadmap**123 124 Prioritize and sequence remediation:125 126 **Priority 1 (Critical - implement before deployment):**127 - High-risk gaps in critical regulations128 - Gaps affecting consequential decisions (loan, hiring, medical)129 - Gaps in fairness/anti-discrimination controls130 131 **Priority 2 (High - implement within 90 days):**132 - Medium-risk gaps in core regulations133 - Gaps in documentation/transparency134 - Gaps in monitoring procedures135 136 **Priority 3 (Medium - implement within 6 months):**137 - Low-risk gaps138 - Gaps in less-critical regulations139 - Enhancements beyond compliance baseline1401416. **Document Regulatory Assessment**142 143 Prepare Regulatory Mapping document including:144 - Applicable regulations identified145 - Regulatory requirements mapped to controls146 - Compliance assessment (current state)147 - Gaps and risk assessment148 - Remediation plan with timeline149 - Sign-off by Legal and Compliance150151## Example152153**Scenario:** US fintech company deploying AI model to predict loan defaults. Model used for pre-approval screening. Model affects both consumer loan applicants and business loan decisions.154155```156# Regulatory Mapping: Loan Default Prediction Model157158## 1. System Scope159160**Model:** Loan Default Risk Predictor (LDP) 161**Business Purpose:** Pre-approval scoring for consumer auto loans and small business lines of credit 162**Geographic Scope:** All 50 US states + DC; some international customers (secondary market) 163**Affected Parties:** Consumer loan applicants (primary; FCRA protected); business applicants (secondary; less regulated) 164**Decision Type:** Consequential (affects loan approval/denial/interest rate) 165**Data Scope:** Credit bureau data, bank account data, transaction data 166167---168169## 2. Applicable Regulations - Identified170171### Primary (Direct Impact)172173**Federal Fair Lending Laws:**174- Fair Credit Reporting Act (FCRA, 15 U.S.C. § 1681)175- Equal Credit Opportunity Act (ECOA, 15 U.S.C. § 1691)176- Fair Housing Act (FHA, 42 U.S.C. § 3601) - if used for mortgage or housing-related credit177- Community Reinvestment Act (CRA) - if depository institution178179**Consumer Privacy Laws:**180- Gramm-Leach-Bliley Act (GLBA) - financial privacy rules181- California Consumer Privacy Act (CCPA) - if California customers182- State privacy laws (NY BitLicense, MA requirements, etc.)183184**AI/Algorithmic Transparency Laws:**185- NY Local Law 144 (algorithmic accountability; requires notice that AI used in decisions)186- State consumer protection laws (FTC Act enforced by states)187- NIST AI Risk Management Framework (guidance, not binding but increasingly expected)188189**Industry Guidance:**190- FDIC/Fed/OCC Fair Lending Guidance (2023)191- EEOC AI Guidance (emerging standards for employment; not directly applicable but principles informative)192193### Secondary (Indirect)194195**Data Security:**196- State data breach notification laws197- PCI DSS (if credit card data handled)198- FTC Safeguards Rule (for financial institutions)199200**International (if applicable):**201- GDPR (if EU customers; less likely for consumer loans but assess)202- UK AI Bill (if UK expansion planned)203204---205206## 3. Regulatory Requirements Mapped to Controls207208### FCRA Compliance (Fair Credit Reporting Act)209210| FCRA Requirement | Application to LDP Model | Operational Control Required | Current Status | Gap |211|---|---|---|---|---|212| **Permissible Purpose** | Model must be used ONLY for credit decisions (loan/credit limit/interest rate). Cannot use for marketing, employment, insurance, etc. | Document model's permitted use case; restrict model API access to loan decisions only | Yes — documented in model card | None |213| **Accuracy** | Ensure model training data is accurate; monitor for systematic errors | Validate training data sources; quarterly accuracy audits; retraining pipeline | Partial — accuracy audit not currently scheduled | Medium: Implement quarterly audit schedule |214| **Adverse Action Notice** | When loan is DENIED (or credit terms unfavorable), provide consumer with notice explaining the reason and right to dispute | When model score triggers decline, system must provide adverse action notice (not just model score) | Partial — adverse action sent but insufficient explanation of model factors | Medium: Enhance notice to explain top factors influencing score |215| **Dispute Procedures** | Consumer can dispute accuracy of data/decision; company must reinvestigate | Establish process for loan applicants to dispute model decisions; escalate to human review | No — not currently implemented | High: Critical for compliance; implement dispute intake form and process |216| **Non-Discrimination** | Model cannot discriminate based on protected class (race, color, religion, national origin, sex, marital/family status, age, receipt of public assistance) | Monitor outcomes by protected class; conduct disparate impact analysis; identify proxy variables | Partial — demographic monitoring attempted but incomplete; no disparate impact analysis | High: Implement comprehensive disparate impact testing |217| **Fair Credit Reporting Practices** | Follow FTC guidelines for consumer information use | Privacy notice (privacy policy), consent practices, data security | Yes — privacy policy and consent in place | None |218219**FCRA Assessment:** Partial compliance. Critical gaps in dispute procedures and disparate impact analysis.220221---222223### ECOA Compliance (Equal Credit Opportunity Act)224225| ECOA Requirement | Application to LDP Model | Control | Status | Gap |226|---|---|---|---|---|227| **Non-discrimination** | Cannot discriminate based on sex, marital status, age, race, color, religion, national origin, receipt of public assistance, or exercise of consumer rights | Ensure model not trained on protected class features; monitor for proxy discrimination | Partial — no protected class features used explicitly, but proxy risk assessed | Medium: Document proxy variable risk; implement monitoring |228| **Adverse Action Notice** | Similar to FCRA; consumer right to know why credit was denied | See FCRA adverse action notice requirement | Partial | Medium: (same as FCRA) |229| **Appraisal Independence** (if mortgage-related) | Independent appraisals required; model cannot override | Not applicable to auto/business loans | N/A | None |230231**ECOA Assessment:** Largely aligned with FCRA compliance. Focus on disparate impact and proxy discrimination.232233---234235### NY Local Law 144 (Algorithmic Accountability; if customers in NY)236237| Requirement | Application to LDP | Control | Status | Gap |238|---|---|---|---|---|239| **Notice** | Notify consumers that automated decision system is used in hiring (not directly applicable to credit but principle is important) | For credit decisions, disclose that AI is used; explain how it works | Partial — privacy policy mentions model but not clear or prominent | Low: Add clear disclosure to loan application and adverse action notice |240| **Annual Bias Audit** | Audit for bias and disparate impact; document findings | Conduct and document annual fairness/bias audit | No — not currently audited | High: Critical for compliance; implement annual audit by [Date] |241| **Data Minimization** | Use only necessary data; justify use of each data source | Document why each data source is used; remove unnecessary features | Partial — features justified but not systematically reviewed | Medium: Conduct feature justification review |242243**LL 144 Assessment:** Significant gaps. Not yet compliant; must implement before full deployment.244245---246247### State Consumer Protection / GLBA (Financial Privacy)248249| Requirement | Control | Status | Gap |250|---|---|---|---|251| Privacy notice (what data you collect and how you use it) | Detailed privacy policy; model use disclosed | Yes | None |252| Consent (for some data sharing) | Opt-in consent for sharing with third parties | Yes — privacy policy includes consent mechanism | None |253| Data security (reasonable security measures) | Encryption, access controls, breach notification plan | Partial — encryption in place; access controls limited | Medium: Strengthen access controls |254| Breach notification | Notify affected users if breach occurs | Process documented | None |255256---257258## 4. Compliance Gaps & Risk Assessment259260### High-Risk Gaps (Must remediate before deployment)261262**Gap 1: Disparate Impact Analysis**263- **Regulation:** FCRA, ECOA264- **Requirement:** Monitor outcomes for discrimination; ensure model does not have disparate impact265- **Current State:** No systematic monitoring266- **Risk Level:** High — regulatory enforcement risk, litigation risk, reputational risk267- **Remediation:** 268 1. Conduct baseline disparate impact analysis (identify protected classes; compare outcomes)269 2. Establish quarterly monitoring270 3. Design mitigation if disparate impact detected (e.g., fairness constraints, manual review threshold)271 4. **Timeline:** By [Deployment Date - 4 weeks]; ongoing quarterly272273**Gap 2: Dispute Procedures**274- **Regulation:** FCRA275- **Requirement:** Provide process for consumers to dispute decisions276- **Current State:** No documented process277- **Risk Level:** High — direct FCRA violation; regulatory enforcement likely278- **Remediation:**279 1. Design dispute intake process280 2. Document reinvestigation procedure281 3. Train staff on dispute handling282 4. Log all disputes and outcomes for audit trail283 5. **Timeline:** By [Deployment Date - 2 weeks]; before going live284285**Gap 3: Bias Audit Procedures**286- **Regulation:** NY LL 144, FTC guidance287- **Requirement:** Conduct annual bias audit; document findings288- **Current State:** Audit not scheduled; no procedures defined289- **Risk Level:** Medium (regulatory expectation, not yet legal requirement nationally)290- **Remediation:**291 1. Define audit scope, methodology, protected classes292 2. Schedule annual audit before [Date]293 3. Document findings and remediation actions294 4. **Timeline:** First audit by [Date + 90 days]; then annually295296### Medium-Risk Gaps (Remediate within 90 days of deployment)297298**Gap 4: Enhanced Adverse Action Notices**299- **Regulation:** FCRA300- **Requirement:** Notice must explain decision; currently sent but lacks detail301- **Risk Level:** Medium — currently compliant but opportunity to enhance302- **Remediation:** Update notice template to include top 3 model factors; example: "Your application was not approved primarily due to: (1) High debt-to-income ratio, (2) Limited credit history, (3) Recent delinquency. You have the right to..."303- **Timeline:** Implement by [Deployment + 30 days]304305**Gap 5: Access Control Enhancement**306- **Regulation:** GLBA, GDPR (if EU customers)307- **Requirement:** Restrict data access; log all access308- **Risk Level:** Medium — security and privacy risk309- **Remediation:** Implement field-level access control; log all credit bureau data access310- **Timeline:** Implement by [Deployment + 90 days]311312---313314## 5. Regulatory Compliance Status Summary315316| Regulation | Critical Requirement | Status | Gap Level | Remediation Timeline |317|---|---|---|---|---|318| FCRA | Adverse action notice | Partial | Medium | Deployment + 30 days |319| FCRA | Dispute procedures | No | High | Before deployment |320| FCRA | Accuracy monitoring | Partial | Medium | Q1 Year 2 |321| FCRA | Permissible purpose | Yes | None | — |322| ECOA | Non-discrimination | Partial | High | Before deployment |323| ECOA | Disparate impact analysis | No | High | Before deployment |324| NY LL 144 | Notice of AI use | Partial | Low | Deployment + 30 days |325| NY LL 144 | Annual bias audit | No | Medium | Deployment + 90 days |326| NY LL 144 | Data minimization | Partial | Low | Deployment + 60 days |327| GLBA | Privacy & consent | Yes | None | — |328| GLBA | Data security | Partial | Medium | Deployment + 90 days |329330**Overall Compliance Status:** CONDITIONAL APPROVAL331332- **Before Deployment:** Remediate dispute procedures (Gap 2) and disparate impact analysis (Gap 1)333- **By Deployment + 30 days:** Remediate adverse action notices (Gap 4) and AI use disclosure (LL 144)334- **By Deployment + 90 days:** Remediate bias audit (Gap 3), data minimization (Gap 5), access controls (Gap 5)335336---337338## 6. Regulatory Compliance Roadmap339340```341BEFORE DEPLOYMENT (Critical)342├─ Disparate impact analysis: complete baseline assessment343├─ Dispute procedures: design and document process344├─ Staff training: ensure staff understand FCRA/ECOA requirements345└─ Legal review: final sign-off on compliance posture346347DEPLOYMENT + 30 DAYS (High Priority)348├─ Adverse action notice: implement enhanced template with model factors349├─ AI use disclosure: add clear disclosure to loan app and notices350└─ Monitoring dashboard: implement to track outcomes by protected class351352DEPLOYMENT + 90 DAYS (Medium Priority)353├─ Bias audit procedures: define and document annual audit process354├─ Access control: implement field-level restrictions and logging355├─ Data minimization: review features; document justification for each356└─ Incident response: finalize procedures for regulatory inquiries357358ONGOING (Continuous)359├─ Quarterly: disparate impact monitoring (protected class outcomes)360├─ Quarterly: adverse action audit (sample 10 denials; verify notice provided)361├─ Annually: bias and fairness audit (full assessment)362├─ Continuously: dispute intake and reinvestigation (log all)363└─ Continuously: incident response (escalate any signs of discrimination)364```365366---367368## 7. Approvals & Sign-Off369370**Regulatory Compliance Assessment completed:** [Date]371372- **Legal Counsel:** _____________________ Date: _____ [Confirms regulatory assessment and remediation plan]373- **Compliance Officer:** _____________________ Date: _____ [Confirms compliance roadmap feasible and resourced]374- **Model Owner:** _____________________ Date: _____ [Confirms ability to implement technical controls]375- **Executive Sponsor:** _____________________ Date: _____ [Approves deployment contingent on Gap 1 & 2 remediation]376377**Deployment Authorized:** [Date] pending completion of Critical remediation items378379**Follow-up Audit Scheduled:** [Date + 120 days]380381---382383## Regulatory Reference384385**Federal:**386- FCRA: 15 U.S.C. § 1681387- ECOA: 15 U.S.C. § 1691388- GLBA: 15 U.S.C. § 6801389- FDIC/Fed/OCC Fair Lending Guidance (2023)390391**State:**392- NY LL 144: Local Law 144 (2023)393- CA CCPA: California Consumer Privacy Act (2018)394- MA AI Transparency: "An Act Relative to the Oversight of Artificial Intelligence" (2023)395396**International:**397- GDPR: Regulation (EU) 2016/679398- UK AI Bill (pending)399400**Guidance:**401- NIST AI Risk Management Framework (2023)402- FTC "Endorsement Guides" for AI recommendations403```404405---406407## Regulatory Mapping Checklist408409Ensure comprehensive assessment:410411**Regulatory Identification:**412- [ ] Geographic scope identified (US/EU/other jurisdictions)413- [ ] Industry identified (finance, healthcare, employment, consumer)414- [ ] Decision type identified (consequential, ranking, advisory)415- [ ] All applicable regulations identified (federal, state, industry-specific)416- [ ] Regulatory landscape documented417418**Requirements Mapping:**419- [ ] For each regulation, specific requirements identified420- [ ] Requirements mapped to operational controls421- [ ] Control implementation status assessed (yes/no/partial)422- [ ] Evidence of compliance documented423424**Gap Assessment:**425- [ ] Current state vs. required state documented426- [ ] Risk level assigned to each gap (critical/high/medium/low)427- [ ] Remediation required or waived428- [ ] Resource requirements estimated429430**Compliance Roadmap:**431- [ ] Remediation prioritized (before deployment vs. after)432- [ ] Timeline established with specific dates433- [ ] Owner assigned for each remediation434- [ ] Dependencies identified (order of implementation)435436**Documentation & Approval:**437- [ ] Regulatory Mapping document prepared438- [ ] Gap assessment documented439- [ ] Compliance roadmap documented440- [ ] Sign-off from Legal, Compliance, and Executive Sponsor441- [ ] Follow-up audit scheduled