OpenAI pulls its Scarlett Johansson-like voice for ChatGPT

OpenAI is pulling the ChatGPT voice that sounds remarkably similar to Scarlett Johansson after numerous headlines (and even Saturday Night Live) noted the similarity. The voice, known as Sky, is now being put on “pause,” the company says.

“We believe that AI voices should not deliberately mimic a celebrity’s distinctive voice— Sky’s voice is not an imitation of Scarlett Johansson but belongs to a different professional actress using her own natural speaking voice,” OpenAI wrote this morning.

OpenAI CTO Mira Murati denied that the imitation of Johansson was intentional in an interview with The Verge last week. Even if Johansson’s voice wasn’t directly referenced, OpenAI CEO Sam Altman was seemingly already aware of the similarities, posting the single-word message “her” on X after the event. Altman has previously said that Spike Jonze’s Her, which features Scarlett Johansson voicing a sultry-sounding virtual assistant, is his favorite movie.

OpenAI did not mention if it had received any contacts about possible legal issues or challenges regarding its assistant’s similarities with Johansson or the role she plays in Her. The Verge has reached out to a representative for Johansson for comment.

ChatGPT’s voice mode and the Sky voice model have been around since last year. But the feature was made far more prominent last week, when OpenAI demoed advancements it made as part of its new GPT-4o model. The new model makes the voice assistant more expressive and allows it to read facial expressions through a phone’s camera and translate spoken language in real time.

The five currently available ChatGPT voice profiles were selected from over 400 casting submissions from voice and screen actors, according to OpenAI. The company declined to share the names of the actors, citing the need to “protect their privacy.”

The new ChatGPT voice assistant capabilities will launch “in the coming weeks” as a limited “alpha” release for ChatGPT Plus subscribers. OpenAI plans to eventually introduce additional voices to “better match the diverse interests and preferences of users.”