Bias Audit
Quick Definition
A bias audit is an independent statistical analysis of an automated employment decision tool's outcomes, evaluating whether it produces disparate selection rates across sex, race, and ethnicity categories, required in specific circumstances such as New York City's Local Law 144 before the tool can be used to screen candidates.
What Is Bias Audit?
A bias audit, in the specific sense required under NYC Local Law 144, calculates the tool's selection or scoring rate for each sex, race, and ethnicity category the applicant pool contains, then compares those rates using an impact ratio analysis similar to the 'four-fifths rule' historically used in disparate impact analysis under US employment law. The audit must be conducted by an independent auditor — not the vendor who built the tool and not the employer using it — within one year of the tool's use, and the results summary must be published publicly before the tool is deployed for NYC candidates.
The independence requirement is a meaningful design feature of the audit, not a formality: an auditor with a financial or reporting relationship to the tool's vendor cannot conduct a compliant audit, since the entire purpose of the requirement is to produce a credible, arm's-length check on the vendor's own claims about the tool's fairness. Employers evaluating a vendor's self-reported 'bias-free' or 'audited' claims should confirm the audit meets this independence standard specifically, not simply that some form of internal review occurred.
A bias audit is not a one-time compliance event. Because a tool's underlying model, training data, or scoring logic can change over time — through vendor updates, retraining, or configuration changes made by the employer — laws requiring a bias audit generally expect it to be repeated periodically, at minimum annually under NYC's framework, rather than treated as a single launch-time checkbox that remains valid indefinitely.
Beyond the specific NYC requirement, a bias audit is also a genuinely useful risk-management practice independent of any legal mandate — an employer using an AI screening tool without ever having its actual selection outcomes statistically reviewed has no real basis for knowing whether the tool is producing a defensible, non-discriminatory result, regardless of what the vendor's marketing materials claim.
Why Bias Audit Matters
A bias audit is the concrete mechanism that turns a vendor's marketing claim of 'fair AI' into a verifiable, independently produced statistical result — without one, an employer has no real evidence to defend an AI hiring tool's outcomes if they're ever challenged.
Key Benefits
- Produces independently verified evidence of a hiring tool's actual selection outcomes across demographic groups
- Satisfies the specific legal requirement for deploying an AEDT under laws like NYC Local Law 144
- Surfaces disparate impact issues before they compound across thousands of hiring decisions
- Creates a defensible compliance record distinct from a vendor's own self-reported fairness claims
- Builds a repeatable process for re-auditing a tool as its underlying model or configuration changes
- Strengthens candidate and public trust in an organization's use of AI hiring technology
Common Use Cases
Frequently Asked Questions
What is a bias audit in AI hiring?
Who can conduct a bias audit?
How often does a bias audit need to be conducted?
What method does a bias audit use to evaluate fairness?
InCruiter Products Related to Bias Audit