Bias Audit
Independent assessment of an AI system for discriminatory impact, required under NYC Local Law 144.
How the laws define this
Central to these rules is that a bias audit must be an impartial evaluation conducted by an independent auditor to test an automated tool for disparate impact. The key area of divergence lies in the scope of protected categories evaluated by the auditor. Under New York City Local Law 144 - Automated Employment Decision Tools, testing is tailored strictly to federal Component 1 categories—specifically race, ethnicity, and sex—along with their intersectional categories, requiring detailed calculations of selection rates, scoring rates, and impact ratios. Conversely, NYC Local Law 28 — AI Tenant Screening Restrictions expands the required assessment to cover a broader set of protected characteristics under local, state, and federal anti-discrimination laws, explicitly extending the audit beyond race and gender to evaluate impact based on religion, national origin, disability, and familial status.
Synthesised from the 3 statutory definitions below. Each one is quoted in full, with its source.
Definitions (3)
An impartial evaluation performed by an independent auditor that tests an AEDT for disparate impact on Component 1 categories (race, ethnicity, and sex) and their intersectional categories, and includes calculations of selection or scoring rates and impact ratios across those categories. The audit must report these metrics and justify any excluded categories under the rules.
A bias audit is an impartial evaluation conducted by an independent auditor, including testing an AEDT to assess its disparate impact on persons of any component 1 category (race, ethnicity, and sex) as required by federal law. It aims to identify and mitigate discriminatory outcomes.
An impartial evaluation by an independent auditor to assess an ADT for disparate impact on individuals based on protected characteristics under local, state, and federal anti-discrimination laws, such as race, gender, religion, national origin, disability, and familial status.
Related Terms
Automated Employment Decision Tool (AEDT)
AI system used in hiring or promotion that requires bias audits under NYC Local Law 144....
Disparate Impact
Discriminatory effect of AI systems that disproportionately affects protected groups, even without discriminatory intent....
Third-Party Audit
Independent evaluation of AI system compliance by an accredited external body rather than the provider itself....