OpenAI Built a Voice That Sounded Like Scarlett Johansson. She Was Not Flattered.

They said it was a coincidence. She said it was not. They pulled it anyway.

In May 2024, OpenAI unveiled a new voice for ChatGPT called Sky.

Scarlett Johansson heard it and contacted her lawyers.

"When I heard the released demo," she said in a statement, "I was shocked, angered and in disbelief that Mr Altman would pursue a voice that sounded so eerily similar to mine."

OpenAI denied it. Sam Altman, the CEO, posted a single word on social media: her. A reference to the 2013 film in which Johansson voiced an AI assistant. Then, within 24 hours and without much further explanation, OpenAI quietly pulled the Sky voice offline.

They said it was a coincidence. The timing suggested otherwise.

OpenAI later stated that Sky was voiced by a different actress who had been cast before any approach was made to Johansson, and that the similarity was unintentional. They also confirmed that they had approached Johansson to be the voice of ChatGPT, that she had declined, and that they had reached back out to her agent shortly before the launch.

She said no. They launched Sky anyway. Then they pulled it when she complained.

The legal question is genuinely interesting. Under current US law, using someone's voice without permission can constitute a violation of their right of publicity, but only if the voice is recognisably theirs. A voice that merely sounds similar is a much harder case to make, especially when the company has a different actress claiming it as her own.

But the legal question almost misses the point.

What OpenAI actually did, whether intentionally or not, was build a product designed to make people feel they were talking to someone warm, intelligent, and distinctively human. And the voice they chose sounded, to many ears, exactly like one of the most recognisable voices in Hollywood.

The "her" post from Altman made it worse. It read less like a coincidence and more like a wink.

Johansson's broader point, beyond the legal dispute, is the right one. People with power and resources are making decisions about whose likeness, whose voice, and whose identity gets used in the AI era. And the people whose voices are being used often find out after the fact, if at all.

The Sky voice has not returned. OpenAI has not acknowledged any wrongdoing. No lawsuit was filed. Johansson's statement made clear she found the situation "threatening and disturbing," not just personally but as a question of what precedent it sets.

A few US states have since moved to strengthen voice and likeness protections. The UK has not yet passed equivalent legislation. The EU AI Act includes provisions around voice cloning but enforcement is still being worked out.

In the meantime, the next AI assistant you speak to might sound familiar. Whether or not anyone asked the original first is, currently, a question of ethics rather than law.