3 min read
AI CV screening gets sold on speed – cutting review time from hours to minutes. What gets undersold is everything that can go wrong if the underlying model wasn’t built or audited carefully, and how much of that risk lands on HR, not the vendor, when something goes sideways.
AI CV screening bias is more common than vendors admit
Models trained on historical hiring data can quietly learn to replicate past bias, favoring candidates who resemble previously successful hires rather than evaluating skills on their own merit. This isn’t a hypothetical; it’s been documented publicly in high-profile cases at major companies, which is exactly why regulators have started paying attention.
What compliance actually requires now
Depending on where you operate, you may be legally required to conduct bias audits on automated hiring tools, disclose their use to candidates, or both. New York City’s Local Law 144 was an early example, and more jurisdictions are following with similar automated employment decision tool requirements. Ignoring this isn’t a minor oversight; it’s a legal exposure that grows with hiring volume.
Questions to ask before rolling out a screening tool
- Has the model been audited for adverse impact by an independent third party, and can the vendor share that report?
- How does the tool handle non-traditional resumes – career gaps, non-linear paths, military experience, or self-taught skills?
- Can a human reviewer see and override the AI’s reasoning for a specific candidate, or is it a black-box score?
What accuracy actually means here
Accuracy isn’t just “did it find qualified candidates.” It’s whether it consistently surfaces qualified candidates across different demographic groups at similar rates. A tool that’s 95% accurate for one group and 70% for another has an accuracy problem regardless of the headline number vendors advertise.
A reasonable rollout approach
Run the tool in parallel with human screening for a full hiring cycle before relying on it solely. Compare who the AI flagged versus who a human recruiter would have advanced, and investigate any meaningful gaps before scaling usage further.
Frequently asked questions
Are HR teams legally liable for biased AI screening?
Increasingly, yes, depending on jurisdiction. Using a vendor’s tool doesn’t automatically transfer liability away from the employer using it to make hiring decisions.
Should candidates be told AI is screening their application?
In a growing number of jurisdictions, disclosure is legally required. Even where it isn’t, it’s increasingly considered a best practice for candidate trust.
Full reviews of AI screening tools, including how each handles bias auditing, are in our AI tools directory.