For its part, OpenAI has warned that its product shouldn’t be used for hiring decisions — or any procedure that could violate someone’s human rights.

The latest findings are in line with several recent studies that show AI can give deferential treatment to candidates of certain demographics. Most experts think these biases are embedded in the vast corpus of material used to train AI.

OpenAI and its competitors have said they’re working to fix the problem. In the meantime, recruiters can try putting their own guardrails in place — like removing names before running applications through AI platforms.