Center for Practical AI
Education · AI and Employment

The algorithm that read your resume — and put it in the bin.

75% of resumes submitted to large employers are rejected by software before a human ever sees them. The tools doing the rejecting were often trained on historical hiring data — which encodes decades of documented racial, gender, and economic discrimination. In 2018, Amazon scrapped an AI hiring tool after discovering it systematically downgraded resumes from women. Most companies never discover this, or don't disclose it.

75%

The number behind this guide

of resumes are rejected before a human reads them.

The software doing the rejecting was often trained on historical hiring data — which encodes decades of documented discrimination.

How Hiring AI Works

Before a human reads your resume.

AI hiring tools are embedded across the hiring funnel — from initial screening to video interview scoring to reference checking.

Resumes rejected by ATS before human review at large employers

Fortune 500 companies using ATS

Hiring managers who report ATS filtered out a qualified candidate

Callback gap between Black-perceived and white-perceived names (Kline et al., 2022)

Stage 1: ATS Screening

Workday, Greenhouse, Lever, iCIMS

What it does

Parses resume for keywords, credentials, experience dates, and employer names. Applies hard filters (degree requirement, years of experience). Ranks remaining candidates by match score.

Bias pathway

Trained on which resumes produced hires at that company — inheriting historical hiring preferences. Keyword filtering disadvantages candidates from different industry vocabulary backgrounds.

Stage 2: AI Resume Scoring

HireEz (formerly Hiretual), Eightfold.ai, SeekOut

What it does

AI ranks resumes against ideal profiles built from top performer data. Surfaces candidates from similar employer pedigrees, education institutions, and career trajectories.

Bias pathway

Top performer profiles encode the characteristics of whoever was historically considered a top performer — often filtered by who got opportunities, not who performed.

Stage 3: AI Video Interview

HireVue, Spark Hire, Outmatch

What it does

Candidates complete a video interview with no human present. AI analyzes facial expressions, vocal tone, word choice, vocabulary complexity, and eye contact patterns. Generates a 'fit score.'

Bias pathway

Validated on historical hire data. Documented performance gaps for darker skin tones and non-standard accents (MIT Media Lab). Not meaningfully audited by most employers before deployment.

Stage 4: Background Check AI

Checkr, Sterling, HireRight

What it does

Automated criminal background check flagging. Some systems apply predictive tools to assess risk from record type, date, and context.

Bias pathway

Black Americans are arrested and convicted at higher rates for the same conduct. Automated background check flagging without individualized review replicates this disparity. EEOC guidance requires individualized assessment before adverse action.

Documented Cases

When the algorithm discriminates.

These are not hypothetical. They are documented cases with legal filings, internal investigations, or independent audits.

Amazon — Scrapped AI Hiring Tool (2018)

Outcome: Amazon scrapped the tool after internal review. The company stated the tool was never used for actual hiring decisions — but the case is now the canonical example of training data bias in AI hiring.

iTutorGroup — EEOC Settlement (2023)

Outcome: First EEOC lawsuit and settlement involving AI hiring discrimination. iTutorGroup paid $365,000 in back pay and damages. The EEOC issued guidance that using AI to make age-based employment decisions violates the Age Discrimination in Employment Act regardless of whether humans were involved in the decision.

Workday — Class Action (2023)

Outcome: Case survived initial motion to dismiss (2024). If Workday is found liable as a third-party tool provider — not just the employers who use its software — the precedent would transform accountability for AI hiring tools across the industry.
Algorithmic Management

After you're hired. The algorithm watches.

AI surveillance and management tools monitor worker performance in real time — and can terminate employment without human review.

Amazon warehouse workers monitored by AI for productivity metrics (Time, 2021)

Average productivity target cycle for Amazon picker rates

Injury rate at Amazon vs. industry average — linked to productivity surveillance (AFL-CIO)

Human managers reviewing most AI-generated termination recommendations at Amazon (Reuters, 2021)

Keystroke and screen monitoring

Hubstaff, ActivTrak, Teramind

Used in remote work — captures keystrokes, screenshots, mouse movement. Disproportionately deployed for lower-wage knowledge workers. Creates constant surveillance pressure that research links to burnout and productivity loss.

AI-generated productivity scores

Amazon UPT, TikTok/ByteDance systems, call center AI

Workers receive an algorithm-generated productivity score that affects scheduling, pay, and termination. The scoring criteria may not be disclosed, and workers often have no appeal path.

Emotional and expression analysis

Call center AI (Cogito, NICE, Genesys), customer-facing workers

AI listens to customer calls and provides real-time coaching — 'smile more,' 'slow down,' 'sounds stressed.' Research shows these tools perform worse on workers of color and non-native English speakers.

Route and movement optimization

UPS ORION, Uber/Lyft routing, gig delivery platforms

Productivity baselines are set algorithmically, without accounting for traffic, weather, worker disability, or regional differences. Workers who deviate from the route — for safety or efficiency — are flagged.

The Gig Economy

Independent contractor controlled by algorithm.

Gig platforms claim workers are independent contractors — but the algorithmic control they exercise is more granular than most traditional employment.

Uber and Lyft — algorithmic control and deactivation

  • Uber and Lyft drivers can be 'deactivated' (fired) based on algorithmic assessment of their ratings, cancellation rates, and acceptance rates — without human review.
  • Drivers have documented cases of deactivation following false accusations, facial recognition failures in identity verification, and rating manipulation by passengers.
  • The EEOC has found that algorithmic deactivation without appeal violates worker rights in specific contexts — but gig workers classified as independent contractors have fewer protections than employees.
  • Uber's facial recognition identity verification has documented accuracy gaps for Black drivers — who have been deactivated after the system failed to verify their identity.

The misclassification problem

Platforms classify workers as independent contractors to avoid employment law protections — minimum wage, overtime, anti-discrimination law, workers' compensation. But the algorithmic control they exercise (setting wages, controlling routing, defining performance standards, deactivating for non-compliance) mirrors employer control. Courts and legislators are split on how to classify this. California AB5, the EU's Platform Work Directive, and the DOL's 2024 independent contractor rule are all attempts to address the gap.

Interactive Tool

What does an ATS actually flag
on a resume?

Walk through an annotated resume — section by section — to see what ATS systems flag, where bias enters, and what you can do. Includes facilitator mode for classroom use.

Open the ATS Scanner →

Also see: Algorithmic Bias for the broader context on AI discrimination in criminal justice, housing, and credit.

What You Can Do

Action for every level of influence.

1

For yourself as a job seeker

  • Use a single-column, plaintext-friendly resume format for any position with ATS submission — visual design features cause parsing failures.
  • Mirror the exact language of the job description for required skills. ATS keyword matching is lexical — synonyms are not recognized.
  • If asked to complete an AI video interview, you can request a human interview in lieu. Ask HR explicitly: 'Is AI analysis used to score this interview? I'd like to request a human review.'
2

For workers subject to algorithmic management

  • Request access to the data used to evaluate your performance. Under GDPR (EU), CCPA (California), and VCDPA (Virginia), you have the right to know what data is being collected and how it is used in decisions about you.
  • If you believe an AI tool has made a discriminatory employment decision, contact the EEOC or your state employment discrimination agency. The complaint process is free.
  • Organize with coworkers. Algorithmic management affects everyone in a workplace — collective action on data practices is more effective than individual complaints.
3

For HR and employers

  • Before deploying any AI hiring tool, require a bias audit by an independent third party. NYC Local Law 144 requires this and is a model for compliance.
  • Remove degree requirements that are not genuinely necessary for job performance. These exclude 30–40% of otherwise-qualified workers and disproportionately affect Black and Hispanic candidates.
  • Audit your ATS rejection rate by zip code, institution type, and employment gap. Disparate rejection rates without business justification are EEOC liability.
4

For policy

  • Support the Algorithmic Accountability Act (federal) — would require impact assessments for automated decision systems used in employment.
  • NYC Local Law 144 (2023) is the first US law requiring bias audits for AI hiring tools. Support equivalent legislation in your state or city.
  • Advocate for EEOC enforcement of existing Title VII disparate impact theory as applied to AI — the legal authority exists, enforcement resources are the limiting factor.

For Educators

Teaching AI and employment rights?

Facilitation guide for the ATS scanner, discussion questions on disparate impact, and an employment discrimination advocacy exercise.

Educator Guide →

Want CPAI to deliver worker rights training to your community?

We partner with unions, workforce development programs, legal aid organizations, and community colleges.