Algorithmic Discrimination

Algorithmic discrimination happens when an automated or AI-driven system produces outcomes that disadvantage people based on a protected characteristic — race, sex, age, disability, or national origin — even though no person intended that result. It differs from ordinary human bias mainly in scale: one flawed model repeats the same skewed decision across every applicant it screens, not just one manager's calls.

Existing employment law does not carve out an exception for software. If a tool a company buys or builds produces a discriminatory outcome, the employer using it is still exposed, and regulators like New York City's Department of Consumer and Worker Protection now require proof the tool was checked before it went live.

Who Does Algorithmic Discrimination Law Apply To?

Any employer, staffing agency, or HR platform that uses an automated or AI-assisted tool to screen, rank, hire, promote, discipline, or terminate employees falls under this umbrella. Coverage extends to resume-parsing software, chatbot-based interview scoring, and AI scheduling tools that indirectly affect pay or hours.

Some rules, including Colorado's AI Act, place obligations on both the employer deploying the tool and the vendor that built it, so vendor contracts increasingly need to spell out who runs the bias testing.

What Are the Main Types of Algorithmic Discrimination?

Three patterns show up most often in HR tools, and they require different fixes.

How Should HR Reduce Algorithmic Discrimination Risk?

Run a bias audit before a tool goes live and on a recurring schedule after, comparing selection rates across groups against the standard four-fifths rule.

• Require vendors to share audit methodology and results before signing, not after
• Keep a documented human reviewer in the loop before any adverse action
• Re-test after every model retrain, not just once a year
• Train recruiters on how and when to override an algorithmic score

Why Does Algorithmic Discrimination Matter for HR Right Now?

AI-assisted screening has moved from a handful of large employers to a mainstream part of hiring, which means the exposure isn't theoretical anymore. A tool bought off the shelf, with no visibility into how it was trained, can quietly produce the same skewed outcome across thousands of applications before anyone notices.

HR teams that treat bias testing as a one-time vendor selection question, rather than an ongoing practice, are the ones most likely to be caught off guard when a regulator, journalist, or rejected candidate asks for the audit results.

Discover how our HR solutions streamline onboarding, boost employee engagement, and simplify HR management

Request a Demo

Frequently Asked Questions

Q: Is algorithmic discrimination illegal?

A: Yes. It's assessed under the same anti-discrimination statutes as any other employment decision, whether a person or a model made the call.

Q: Who is liable, the employer or the vendor?

A: Often both. Employers are responsible for how a tool is used; several newer state laws also place testing obligations directly on vendors.

Q: What is a bias audit?

A: An independent statistical comparison of selection rates across protected groups, typically measured against the four-fifths rule.

Q: Does removing protected-class fields from the data prevent it?

A: No. Proxy variables that correlate with a protected trait can reintroduce the same pattern.

Q: How often should audits run?

A: At minimum annually, and again after any meaningful change to the underlying model, per frameworks like the Colorado AI Act.

Q: Can a candidate sue over an algorithmic hiring decision?

A: Yes, under the same anti-discrimination laws that apply to any hiring decision, algorithmic or not.