AIGovernance

Statistical Auditing, Drift Monitoring and Governance Reporting for AI Decision Systems

Disclaimer: AIGovernance provides statistical auditing and documentation support tools only. It does not provide legal advice and does not certify compliance with any law or regulation.

Motivation

Organisations deploying AI systems in employment and public decisions face growing regulatory requirements — from the EEOC 4/5ths adverse impact rule to NYC Local Law 144 (mandatory annual bias audits for AEDTs) to the EU AI Act (High Risk classification for employment AI) and federal requirements for ongoing monitoring of high-impact AI uses.

Existing R fairness packages (for example fairmodels and fairness) measure disparities at a single point in time. Regulatory audits such as the 4/5ths rule and the LL144 impact ratio are also single-period comparisons. They cannot tell whether a change in selection rates comes from a different applicant pool or from a change in how comparable applicants are decided.

AIGovernance combines regulatory-format reporting with auditing of decision drift over time: case mix versus allocation versus rule change, net versus masked (offsetting) drift, and unit-level drift across recruiters, offices or courts.

Frameworks Covered

Framework	Jurisdiction	Coverage
EEOC Uniform Guidelines	US Federal	Adverse impact (4/5ths rule), Fisher & Z tests
NYC Local Law 144	New York City	Impact ratio table, disclosure format, procedural checklist
NIST AI RMF 1.0	US (voluntary)	GOVERN / MAP / MEASURE / MANAGE scoring
EU AI Act	European Union	Risk tier classification, Annex III, key obligations

Installation

# From GitHub (development version)
remotes::install_github("subirhait/AIGovernance")

# From CRAN
install.packages("AIGovernance")

Quick Start

library(AIGovernance)

# --- 1. Built-in synthetic hiring data ---
data(hiring_sim)

# --- 2. Build governance object ---
gov <- aigov_build(
  data        = hiring_sim,
  outcome     = selected,
  group       = race_ethnicity,
  ref_group   = "White",
  frameworks  = c("EEOC", "NYC_LL144", "NIST_RMF"),
  org_name    = "Acme Corporation",
  system_name = "ResumeAI v1.0"
)

# --- 3. Check applicable laws ---
aigov_scope(gov, domain = "employment", us_state = "NY")

# --- 4. EEOC adverse impact ---
gov <- aigov_adverse_impact(gov)

# --- 5. NYC Local Law 144 ---
gov <- aigov_audit_nyc(gov)

# --- 6. NIST AI RMF ---
gov <- aigov_audit_nist(gov, responses = list(
  GOVERN_1_1 = TRUE, MAP_1_1 = TRUE, MEASURE_1_1 = TRUE
))

# --- 7. EU AI Act risk classification ---
gov <- aigov_classify(gov, domain = "employment",
                       makes_final_decision = TRUE,
                       human_oversight = FALSE)

# --- 8. Generate HTML report ---
aigov_report(gov, format = "html")

Decision drift (v0.2.0)

data(hiring_panel)   # 3 periods, 24 recruiters, reject / review / advance

gov <- aigov_build(hiring_panel, selected, group, ref_group = "A")

# case mix vs allocation vs rule change; net vs masked drift; unit drift
gov <- aigov_drift(gov, period = period,
                   covariates = c("qualification", "degree"),
                   unit = recruiter, baseline = 2024, current = 2026,
                   B = 200, seed = 1)

# categorical decision
gov <- aigov_drift(gov, period = period,
                   covariates = c("qualification", "degree"),
                   decision = decision, focal = "advance", B = 200, seed = 1)

# every period against the baseline, Holm-adjusted
gov <- aigov_monitor(gov, period = period,
                     covariates = c("qualification", "degree"), B = 200)

# simulate known mechanisms for method evaluation
sim <- aigov_simulate_drift(n = 2000, scenario = "offsetting_cases", seed = 1)
attr(sim, "truth")$drift

Core Functions

Function	Description
`aigov_build()`	Construct governance audit object
`aigov_scope()`	Determine applicable frameworks by domain + jurisdiction
`aigov_adverse_impact()`	EEOC 4/5ths rule + statistical tests
`aigov_audit_nyc()`	NYC Local Law 144 impact ratios + checklist
`aigov_audit_nist()`	NIST AI RMF 1.0 GOVERN/MAP/MEASURE/MANAGE scoring
`aigov_classify()`	EU AI Act risk tier + NIST risk tier
`aigov_checklist()`	Display checklist items for any framework
`aigov_report()`	Generate HTML audit report
`aigov_drift()`	Decision drift: case mix / allocation / rule change, net and absolute drift, unit drift
`aigov_monitor()`	Drift across several periods with multiplicity adjustment
`aigov_simulate_drift()`	Simulate decisions under known drift mechanisms

Ecosystem

AIGovernance is designed to work alongside:

Package	Role
`decisionpaths`	Longitudinal decision path construction
`DecisionDrift`	Temporal drift detection in repeated AI decisions
`AIBias`	Longitudinal bias amplification analysis

Planned

aigov_remediate() — actionable remediation suggestions
Colorado AI Act (SB 205) module
Illinois AI Video Interview Act module
Integration with AIBias and DecisionDrift objects

References

EEOC (1978). Uniform Guidelines on Employee Selection Procedures. Federal Register, 43(166), 38295–38309.
New York City Local Law 144 (2021, effective 2023). https://www.nyc.gov/site/dca/about/automated-employment-decision-tools.page
NIST (2023). AI Risk Management Framework (AI RMF 1.0). https://doi.org/10.6028/NIST.AI.100-1
European Parliament and Council (2024). Regulation (EU) 2024/1689 (EU AI Act). https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
Kitagawa, E. M. (1955). Components of a difference between two rates. Journal of the American Statistical Association, 50(272), 1168–1194.

Citation

@Manual{Hait2026AIGovernance,
  title  = {AIGovernance: Statistical Auditing, Drift Monitoring and
             Governance Reporting for AI Decision Systems},
  author = {Hait, Subir},
  year   = {2026},
  note   = {R package version 0.2.0},
  url    = {https://github.com/subirhait/AIGovernance}
}

Author

Subir Hait
Michigan State University
haitsubi@msu.edu
ORCID: 0009-0004-9871-9677