We believe AI-powered screening must be transparent, fair, and accountable. This report details how our AI works, how we test for bias, and how we ensure every candidate is treated equitably.
Last updated: April 2026 | Next audit: July 2026
Six core principles guide every decision our AI makes.
Every risk score comes with a plain-English explanation of exactly which factors contributed and how much weight each received. No black boxes.
Our models are tested quarterly for disparate impact across race, gender, age, and disability status. We publish results publicly and retrain when bias is detected.
AI makes recommendations, humans make decisions. Every adverse action requires human review. Our AI never autonomously rejects a candidate.
We only collect and process data strictly necessary for the screening. Personal data is encrypted at rest and in transit, and deleted per retention schedules.
Candidates can request a full explanation of their risk score, dispute any finding, and have their data deleted. We respond within 5 business days.
We monitor our AI's performance in production for accuracy drift, bias emergence, and edge cases. Anomalies trigger automatic human review.
We aggregate data from court records, employment databases, education registries, and other verified sources. We never scrape social media or use unverified data sources.
Our AI uses probabilistic matching with name, DOB, SSN, and address to ensure we're looking at the right person. Match confidence must exceed 98% before inclusion.
Each finding is analyzed for severity, recency, relevance to the position, and jurisdiction-specific rules. The AI considers over 200 factors in its assessment.
A composite risk score (0-100) is generated with a detailed breakdown showing exactly which factors contributed and their individual weights.
Scores below 70 are automatically flagged for human review. Adverse action recommendations always require human approval before proceeding.
Independent third-party audit conducted by AI Fairness Lab, LLC.
Full audit report available upon request. Contact [email protected] for access.