This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
Documents can settle a legal question. They rarely settle a reputational one.
OpenAI is in that position at the end of a week that began with an apology. On Wednesday, the Washington Post reported that, according to documents and recordings it reviewed, OpenAI hired a different actress for its “Sky” voice through a casting call issued last May, and the Post reported that her agent said neither Johansson nor the film “Her” ever came up. If that holds, it is meaningful evidence. It may not change how this episode is remembered.
What each side has said
OpenAI’s own timeline, updated on May 22, says Sam Altman spoke with Johansson on September 11, 2023 about becoming a sixth ChatGPT voice, and that she declined a week later through her agent. On May 10, three days before the GPT-4o launch, Altman contacted her team again to ask whether she might reconsider. OpenAI says it paused Sky on May 19.
On May 20, Johansson released a statement saying she was “shocked, angered and in disbelief” that Altman would pursue a voice “so eerily similar to mine that my closest friends and news outlets could not tell the difference.” She said she had hired legal counsel, who sent two letters asking OpenAI to detail how Sky was created. She also pointed to something Altman did on launch day: he posted a single word on X, “her.”
Altman’s response the same day: “The voice of Sky is not Scarlett Johansson’s, and it was never intended to resemble hers. We cast the voice actor behind Sky’s voice before any outreach to Ms. Johansson.” He added that OpenAI was “sorry to Ms. Johansson that we didn’t communicate better.”
Intent is disputed. The sequence is not. There was a request, a refusal, a second approach days before launch, a voice many people thought sounded like her, and a post referencing her best-known AI role.
Resemblance is a perception fact
Reputation runs on what people believe they heard. Johansson’s statement makes that point directly: friends, family and the public noticed the similarity before anyone looked at a casting file. Whether an expert or a court agrees is a separate question, answered later and read by far fewer people.
This is the uncomfortable part for any company relying on “the records show.” Records answer did you copy? The public was asking does this sound like her, and did they want it to? Those are different questions, and the second one was answered in the first 48 hours.
The most damaging evidence was the CEO’s own post
Without the word “her,” this is a story about two voices that sound alike, which happens. With it, the strongest support for Johansson’s reading came from OpenAI’s own chief executive, in public, in his own words.
Executives treat posts like that as asides. In a dispute, they become the primary source everyone quotes, and they stay attached to the controversy in search results long after the facts are clarified. Six months after OpenAI’s board crisis, it is a second lesson in how executive visibility cuts both ways.
Identity is now part of the interface
The GPT-4o launch leaned heavily on how human the voice felt: quick, expressive, easy to interrupt. Once a voice carries that much personality, it also carries an identity, and a recognizable identity belongs to somebody.
Johansson framed her statement around “the protection of our own likeness, our own work, our own identities” and called for legislation. For public figures, that widens the reputation question. Beyond what is written about you, it now includes whether a product millions of people talk to sounds like you, and what people assume about your consent when it does.
For communications teams
Two practical lessons. First, if the documentation is on your side, have it ready before launch, not after a pause and a celebrity statement. OpenAI’s detailed timeline arrived two days into the story. Second, every executive post during a launch is part of the launch.
The Post’s reporting may well show that OpenAI did not copy anyone. But the version most people will remember was set on Monday, by the actress who said no.