The results are not uniform. Participants rated the two avatars about the same on trust and credit, with Johan slightly ahead but not significantly so. The payment gap persisted regardless. The researchers also could not fully separate gender from human-likeness, because participants perceived the female avatar as less human. Gender alone therefore cannot be identified as the cause of the gap. Participants' stated attitudes also conflicted with their behavior: 30 of 34 interviewees said gender would not change how they treated an assistant. Details beyond the journal placement and release date are not yet available.
For attorneys, the study matters most as a signal about design. The researchers argue that voice, name, and appearance can trigger the same assumptions people hold about human coworkers, and one industry estimate cited in coverage puts AI agent use at nearly 79% of companies in 2026. Clients deploying avatars or voice-based assistants should review how those systems present themselves, particularly where assistants handle tasks tied to pay, allocation, or evaluation. This is a single small lab experiment, so treat its findings as preliminary, not as a general measure of AI bias. Watch for the full paper and for replications with larger, more diverse samples.