Classet: Third Party Bias Audit

Overview

Classet employs Warden AI for continuous third-party bias monitoring, with audits accessible through the Warden AI Assurance Dashboard. The platform aims to meet the strictest existing legal framework for AEDT and AI-Accelerated hiring tools.

Why Third-Party Bias Auditing?

The initiative provides ongoing visibility into fairness, compliance, and system behavior to help stakeholders understand how the AI operates in practice.

Bias Audit Methodology

Classet uses an inference-based methodology for auditing, which is necessary because:

EEOC surveying is often optional and has low completion rates
The platform must monitor for disparate impact across all applicants, not just survey respondents

The approach uses proxy methods including name-based inference, geocoding/address inference, and Bayesian Improved Surname Geocoding (BISG) to approximate demographic distributions and detect potential adverse impact.

Why Inference Is Considered Acceptable

For auditing only: Inference evaluates whether bias exists in outcomes, not to treat individuals differently
Regulatory recognition: U.S. agencies like the CFPB and EEOC acknowledge these proxy methods when direct demographic data isn't available
Statistical accuracy: These methods detect systemic disparities at aggregate levels, even if imperfect for individuals

Safeguards & Limitations

Cannot be used for individual classification in employment decisions
Organizations should maintain transparency about methods and error rates
Self-reported demographic data remains preferable where available