About 20 seconds of audio can now let AI mimic a specific person closely enough that listeners often think both samples are the same speaker.
Laboratory tests cited by Dartmouth researcher Hany Farid found people detect fake voices no better than a coin flip, helping explain a surge in fraud, political deepfakes and corporate credential-reset scams.
The same realism is also restoring speech for people with ALS and enabling translation in a speaker’s own voice, making the technology both medically valuable and socially dangerous.
Gray-area uses are expanding, including celebrity narration tools and posthumous “griefbots” that replay or generate a dead person’s voice, even as evidence on their psychological effects remains thin.
Rules on ownership and use after death still vary by company, state and country, prompting some users such as Jules Rodriguez to specify that their cloned voice should end when they die.