Realistic AI Voices Raise New Fraud Risks Across Diverse Language Communities

Dundee, 21 September 2026 – Increasingly realistic artificial-intelligence voices may expose some language communities to greater fraud risk because listeners can be more likely to believe a synthetic voice is human when it uses their own underrepresented accent or dialect. The finding adds a behavioural dimension to voice-cloning threats and suggests that generic warnings are less effective than explaining what modern systems can imitate.

Research published in the Journal of Cybersecurity tested whether participants believed audio clips were spoken by people or generated by artificial intelligence. Listeners generally showed a tendency to assume voices were human, but the effect was stronger when synthetic speech matched a regional or non-standard dialect familiar to them. Limited prior exposure to technology speaking that way may reduce suspicion and increase perceived authenticity.