Emergences AI, Inc.
Bias Audit
- Version
- 2026-07-03
- Effective
- July 3, 2026
- Status
- Audit not yet completed
- Most recent audit
- —
Where the law treats our assessment as an automated employment decision tool — for example under New York City Local Law 144 — an independent bias audit must be performed and a summary published. This page is where NeoWork publishes that summary and explains how the audit works.
What a bias audit measures
An independent bias audit examines whether the tool's outcomes differ across demographic groups. Because our assessment produces a score, the audit computes, for each group, the rate at which candidates are scored at or above the selection threshold the employer uses (the “selection rate”), and the ratio of each group's selection rate to that of the highest-scoring group (the “impact ratio”). An impact ratio below 0.80 — the U.S. Equal Employment Opportunity Commission's four-fifths rule of thumb — flags a potential adverse impact for closer review.
Categories audited
- Sex (as reported in the underlying data).
- Race/ethnicity, using the U.S. EEO-1 categories.
- The intersection of sex and race/ethnicity.
Categories with too few records to report reliably are marked as such rather than shown with an unreliable number.
Methodology
- The audit is performed by an independent auditor with no financial interest in the outcome.
- It uses historical outcome data where available and, where it is not, test data representative of the candidate population.
- Selection rate and impact ratio are computed per category using the EEOC four-fifths benchmark, alongside the number of records (the sample) for each group.
- Scoring-threshold sensitivity is checked, since the impact ratio depends on where the employer sets its threshold.
- The full method and data plan are documented for the auditor and counsel and summarized here.
Results
When the independent audit is complete, the impact-ratio results appear in the table below. We do not publish provisional or self-generated figures in their place.
| Category | Selection rate | Impact ratio | Records |
|---|---|---|---|
| Sex | — | — | Pending |
| Race / ethnicity | — | — | Pending |
| Sex × race/ethnicity | — | — | Pending |
What we publish when complete
- The name and independence of the auditor.
- The date of the most recent bias audit.
- The date the tool was first offered for use (distribution date).
- The impact ratios for each category, with the source and number of records used.
- A summary of the results and any resulting changes.
How our design reduces bias in the meantime
- Scoring is based on cited evidence from the work against explicit criteria, not overall impression.
- Protected characteristics and their proxies are excluded from inputs and are never inferred.
- An unobserved behavior is treated as insufficient evidence and flagged for human review, not penalized.
- A human at the employer makes every decision.
For employers and candidates
Employers who need audit data for their own AEDT obligations, and candidates with questions about fairness, can contact us at contact@emergences.ai.