AI System Impact Assessment

AI System Impact Assessment

System Name: [____________________________________]

Organization: [____________________________________]

Assessment Date: [__/__/____]
Next Review Date: [__/__/____]

1. System Overview & Deployment Context

1.1 System Description

What is the system?

[________________________________]

What decisions or recommendations does it make?

[________________________________]

What data inputs does it use?

[________________________________]

Who deployed it (vendor/internal)?

[________________________________]

1.2 Deployment Scope

Geographic jurisdictions where the system is used:

☐ New York City (NYC LL 144 applies)
☐ European Union / EEA (EU AI Act applies)
☐ United States—Federal EEOC jurisdiction
☐ Other: [____________________________________]

Operational context:

☐ Employment hiring/screening
☐ Employment promotion or wage decisions
☐ Layoff or termination decisions
☐ Biometric identification
☐ Access to essential services (public or private)
☐ Law enforcement
☐ Educational/vocational training
☐ Other: [____________________________________]

Frequency of use:

• Expected decision volume per [week/month/year]: [__________]
• First deployment date: [__/__/____]
• Significant model changes since launch: ☐ Yes ☐ No

2. Affected Stakeholder Mapping

2.1 Categories of Affected Persons & Groups

Direct beneficiaries/subjects of decisions:

[________________________________]

Demographic categories at elevated risk:

☐ Race/ethnicity (specify: [____________________])
☐ Sex/gender identity
☐ Age (40+)
☐ Disability status
☐ Religion
☐ National origin
☐ Genetic information
☐ Pregnancy-related status
☐ Intersectional combinations
☐ Other vulnerable groups: [____________________]

Downstream affected parties (e.g., rejected candidates, monitored employees, community members):

[________________________________]

Estimated number of persons affected per cycle:

[__________]

3. Harm Taxonomy & Risk Identification

3.1 Potential Harms

Mark each applicable risk and describe how it may materialize:

Discrimination & Disparate Impact

☐ Discriminatory screening: System rejects candidates disproportionately based on protected characteristic.

Risk description: [________________________________]

Protected classes most at risk: [____________________]

☐ Proxy discrimination: System uses facially neutral data that correlates with protected status (e.g., zip code as proxy for race).

Risk description: [________________________________]

☐ Intersectional bias: Compounded harm affecting individuals in multiple protected categories (e.g., Black women, older disabled workers).

Risk description: [________________________________]

Fundamental Rights Violations (EU FRIA)

☐ Right to non-discrimination: System violates Charter of Fundamental Rights Article 21.

☐ Right to fair trial/due process: System lacks transparency or appeal mechanism.

☐ Right to private/family life: System collects or uses sensitive personal data.

☐ Right to freedom of expression/information: System restricts access to information or opportunities.

☐ Right to work/employment: System unfairly blocks access to employment.

☐ Right to education: System affects access to training or skill development.

☐ Other fundamental rights: [____________________]

Algorithmic Opacity & Autonomy Risks

☐ Lack of interpretability: End-users cannot understand why decisions were made.

☐ Autonomous decision-making: System operates with minimal human oversight.

☐ Feedback loops: Biased historical data reinforces discriminatory patterns.

☐ Adversarial manipulation: System is vulnerable to gaming or evasion.

Accessibility & Accommodation Gaps

☐ Failure to accommodate disability: System does not offer alternative assessment methods (e.g., accessible video interview format).

☐ Accessibility barriers: Interface/process inaccessible to individuals with disabilities.

3.2 Severity & Likelihood Assessment

For each identified risk:

4. Regulatory Compliance Checklist

4.1 NYC Local Law 144 (AEDT Bias Audit)

Applies if: Using automated employment decision tool in NYC hiring/promotion context.

Audit vendor/firm: [____________________________________]
Audit completion date: [__/__/____]
Next audit due: [__/__/____]

4.2 EU AI Act Article 27 (FRIA)

Applies if: Deploying high-risk AI system within EU/EEA (biometric, employment, essential services, law enforcement, migration, justice, etc.).

4.3 EEOC & Federal Anti-Discrimination Laws

Applies to: All AI employment decisions (Title VII, ADEA, ADA, GINA, FCRA).

5. Mitigation Measures & Governance

5.1 Human Oversight & Decision-Making

What human review occurs before consequential decisions?

☐ Individual review of all system outputs
☐ Sampling-based review ([_____]% of decisions)
☐ Review only when system output falls in uncertain range
☐ Review only upon appeal/challenge
☐ No meaningful human review

Human reviewer authority & training:

• Reviewer role/title: [____________________________________]
• Decision authority: ☐ Can override system ☐ Can recommend but not override
• Training on bias & discrimination: ☐ Yes ☐ No | Last date: [__/__/____]
• Documentation requirement: ☐ Yes ☐ No | Format: [____________________]

5.2 Transparency & Explainability

How is the decision communicated to the affected person?

☐ Full explanation of factors/score provided
☐ Limited explanation (top factors only)
☐ System decision only, no explanation
☐ Placeholder text: [____________________________________]

How can an affected person request information about the decision?

• Appeal process: [____________________________________]
• Response timeline: [__________] business days
• Right to human review: ☐ Yes ☐ No
• Right to explanation: ☐ Yes ☐ No

5.3 Data & Model Governance

Training data:

• Data sources: [____________________________________]
• Time period covered: [__________] to [__________]
• Known biases or limitations: [____________________________________]
• Regular validation/testing: ☐ Yes ☐ No | Frequency: [____________________]

Model updates:

• Version control maintained: ☐ Yes ☐ No
• Major changes trigger reassessment: ☐ Yes ☐ No
• Performance monitoring (accuracy, fairness): ☐ Yes ☐ No | Metrics: [____________________]

5.4 Monitoring & Accountability

Ongoing fairness monitoring:

• Selection rates by protected class tracked: ☐ Yes ☐ No
• Performance monitoring schedule: [____________________________]
• Threshold for escalation/intervention: [____________________________]
• Responsible team: [____________________________________]

Complaint & escalation process:

• Internal escalation contact: [____________________________________]
• External escalation (regulatory): ☐ NYC DCWP ☐ EEOC ☐ EU AI Office ☐ Other: [____________________]
• Documentation retention period: [__________] years

6. Accessibility & Accommodation

6.1 ADA / EU Accessibility Requirements

Alternative assessment methods available:

☐ Yes — Specify: [____________________________________]

☐ No — Justification: [____________________________________]

Process for requesting accommodation:

[____________________________________]

Timeframe for accommodation provision: [__________] business days

7. Documentation & Record-Keeping

7.1 Assessment Documentation

Assessment completed by:

• Name: [____________________________________]
• Title/Role: [____________________________________]
• Email: [____________________________________]
• Date: [__/__/____]

Internal legal review:

☐ Yes — Reviewed by: [____________________] | Date: [__/__/____]

☐ No — Justification: [____________________________________]

External legal/compliance review:

☐ Yes — Firm/Consultant: [____________________] | Date: [__/__/____]

☐ No

7.2 Records to Maintain

Required documentation to retain:

☐ This assessment (original & updated versions)
☐ Vendor documentation & claims
☐ Independent bias audit report (NYC LL 144)
☐ FRIA report (EU AI Act)
☐ Disparate impact analyses
☐ Decision logs/audit trail
☐ Human review documentation
☐ Complaint records
☐ Accommodation requests & resolutions

Retention period: [__________] years (per applicable law: NYC LL 144 = 3 years, EEOC = 1+ years, EU = per GDPR/FRIA requirements)

8. Regulatory Notification & External Submissions

8.1 Notifications Required

9. Attestation & Sign-Off

I confirm that this assessment has been completed accurately and that identified risks and mitigation measures have been reviewed and approved for deployment.

Organization Representative:

• Name (Print): [____________________________________]
• Title: [____________________________________]
• Signature: [____________________________________]
  Date: [__/__/____]

Internal Legal/Compliance Officer (if applicable):

• Name (Print): [____________________________________]
• Title: [____________________________________]
• Signature: [____________________________________]
  Date: [__/__/____]

Sources and References

NYC Local Law 144 (AEDT)

• NYC Department of Consumer & Worker Protection — AEDT Guidance
• DCWP-AEDT FAQ & Educational Materials
• 29 CFR § 1607 — Uniform Guidelines on Employee Selection Procedures

EU AI Act Article 27 (FRIA)

• EU AI Act Article 27 — Fundamental Rights Impact Assessment
• CEDPO Micro-Insights: Fundamental Rights Impact Assessments (January 2025)
• European Commission AI Office — AI Act Resources

EEOC Guidance & Federal Anti-Discrimination Laws

• EEOC Enforcement Guidance — Employment Discrimination and AI for Workers (April 2024)
• EEOC Strategic Enforcement Plan FY 2024–2028
• EEOC Technical Guidance on AI & Algorithmic Fairness Initiative
• Title VII of the Civil Rights Act of 1964
• Americans with Disabilities Act (ADA) — AI Tools & Accommodations

Additional Resources

• ACM FAccT 2024 — "Null Compliance: NYC Local Law 144 and the Challenges of Auditing Employment Algorithms"
• Montreal AI Ethics Institute & Purdue GRAIL — AI Policy Corner: NYC Local Law 144