Behind the Scenes of Scarlett Johansson’s Battle With OpenAI

Altman didn’t give up. In mid-May, he texted Lourd, co-chairman of Creative Artists Agency, asking if Johansson might reconsider—he wanted to show the actress something he’d been working on, people familiar with the interaction said. The camps couldn’t settle on a time to meet.

Then on May 13, OpenAI showcased an updated AI system, equipped with new voice assistants for its Chat GPT tool, including a female named Sky. 

Johansson was surprised and angry. She and Lourd thought—and others agreed—that Sky’s voice sounded “eerily similar” to the actress. Lourd and the actress spent the morning fielding calls and emails from friends and associates, some of whom worried that OpenAI had simply appropriated Johansson’s voice without permission.

Tap to listen to OpenAI's 'Sky' voice assistant

Stop

In a 2023 product update, OpenAI published a sample of its voice assistant called "Sky." OpenAI said that the voice was added to ChatGPT and showcased online in a demonstration video on May 13.

When Lourd confronted Altman, however, the OpenAI chief executive was incredulous. Did they really think the voice sounded like Johansson? Was she mad? 

So began the most dramatic episode yet in the collision between Hollywood and the exploding world of artificial intelligence. 

The emergence of AI as a rapidly advancing and perhaps unstoppable force has sparked deep anxiety in creative industries that for decades have been governed by strict rules of how creators are compensated for their work. The reason is that the language models that power generative AI chat tools are typically made using text, images, music and videos hoovered up from across the internet. That can include material that is copyrighted, valuable and often paywalled—like Scarlett Johansson’s voice. 

Johansson—who just three years ago waged a blistering and public legal campaign against Disney—hired a legal team to demand answers from Altman and OpenAI and issued an excoriating statement.

OpenAI, however, said Sky was never intended to resemble Johansson, and that the company had hired a voice actor who recorded the part before any outreach to Johansson. People close to Altman say he wanted Johansson to be involved in the voice project, potentially as an additional voice or to promote the product. 

OpenAI paused use of the Sky voice on Sunday after receiving legal letters from Johansson’s team of representatives. Altman said Monday evening in a statement that he apologized for failing to communicate better. 

OpenAI has spent months making the rounds with studios and producers showcasing its Sora text-to-video tool and discussing potential licensing deals, according to people familiar with the meetings. News Corp, owner of The Wall Street Journal, struck a content-licensing partnership with OpenAI Wednesday that could be worth more than $250 million over five years.

A cash-strapped Hollywood has tiptoed toward generative AI tools, hoping it will save money on tasks involving scripts, production schedules and visual effects. Boosters say AI will speed up mundane tasks, offer payouts to actors who grant rights to AI versions of their voices and could help stars create synthetic doubles of themselves to maximize the number of commercial projects they can pursue at once. Some stars have begun hiring advisers to help them spot instances of their likeness being misused and issue takedown notices.

Yet as talent contemplates what AI means for the future use of their likeness, studios continue to explore opportunities to license content to data-hungry AI companies or build engines that they can use internally. 

Media companies are starting to do their own deal making. Disney is discussing a deal with Microsoft—OpenAI’s strategic partner and biggest investor—for a private generative AI tool that could be trained on Disney’s library of content and other data and used internally, according to people familiar with the matter. The company has also had recent discussions with OpenAI and others. Disney declined to comment. Microsoft declined to comment.

Lawmakers are just starting to take notice of the changing landscape for intellectual property rights. There are bills in Congress that aim to protect artists including the “No Fakes Act” which would prohibit the unauthorized use of digital replicas without consent. Tennessee Gov. Bill Lee signed into law the Ensuring Likeness Voice and Image Securities (ELVIS) Act in March, which makes people’s voices protected personal rights.

Deep fears remain that the very tools that could transform the industry might be able to do so only because they were built using decades of creative work published on the internet without permission or compensation. For many, the Johansson incident was proof positive of how the “move fast and break things” ethos of developing new technologies could erode the cornerstone of Hollywood artistry.

“How these companies align with the actual individuals and creators is what’s key here—the verification of authenticity and receiving consent, and remuneration for consent,” CAA’s Lourd said in a statement. “It’s not too late for these companies to slow down and put processes in place to ensure that the products that are being built are built transparently, ethically, and responsibly.”

Altman has been the most visible face of the AI movement since OpenAI launched ChatGPT in late 2022 and ignited a global frenzy over AI technology. 

He and the company face numerous challenges, including a slew of copyright lawsuits and mounting pressure to advance its GPT-4 technology. It’s also trying to move past its leadership crisis from last November, when OpenAI’s then-board of directors fired Altman for failing to be “consistently candid.” He was quickly reinstated as CEO.

For performers like Johansson and IP owners, it is hard to prove whether their likeness or content has been misused. Regulations governing the systems are scant. 

In the absence of clear rules, OpenAI has said it’s speaking directly to content creators, including studios and artists, about the potential disruption of AI. In interviews, Altman has said the broader AI industry needs to find a way to pay artists for material that is valuable to future AI systems. 

Artists should also be able to opt out of allowing AI systems to mimic their work, he said in a podcast interview in March. 

Altman in a speech last year said he and other OpenAI executives were inspired by the 2013 film “Her,” in which a man falls in love with his AI assistant, voiced by Johansson. The company aimed to develop an assistant like Apple’s Siri or Microsoft’s Cortana that users could talk to.

Last May, the company sent out a casting call looking for male, female and nonbinary voices in the 25 to 45 age range. It wanted voices that were warm, engaging and charismatic, internal documents reviewed by The Wall Street Journal show. “Someone you instantly trust and feel a kinship with. Nothing ‘put on.’” 

It whittled down a list of 400 applicants and flew actors to San Francisco last June and July for recording sessions. The actors were asked to sign nondisclosure agreements and refrain from providing voice recordings to OpenAI’s competitors for three years after the product launch, the documents show. OpenAI says that the actress who played Sky was recording in the studio last July. 

Altman first reached out to Lourd, whom he had known for years personally and professionally, in mid-August and met with Johansson in early September about working with OpenAI, people familiar with the interaction said.

Johansson said in a statement that Altman had told the actress that her voice would be comforting to users and help people feel comfortable with AI’s proliferation.

Few people within OpenAI were aware of Altman’s outreach to Johansson, including some who worked directly on the project, people familiar with the matter said. 

The outreach came months into a Hollywood actors’ strike that shut down production, upending release calendars and putting performers out of work. Concerns over studios’ use of AI and the potential for their likeness to be used without consent or compensation were among the issues at the center of the strike. 

Johansson considered the opportunity and told Altman directly that she wasn’t interested, the people said. The actress declined to work with AI for personal reasons, she said in a statement this week. 

OpenAI debuted the voice capability in late September, but the product largely flew under the radar until this month, when the company released a product demo online that featured the Sky voice. 

On Monday May 13, the day OpenAI released an updated version of its flagship AI system called GPT-4o, Altman posted the word “her” on X. Emails to the actress from friends and associates streamed in asking if she’d participated in the OpenAI project.

Johansson quickly found herself at the center of the battle over what power artists have over the use of their likeness in the age of generative AI. 

“There is seemingly a sense of tone deafness toward artists, creatives and intellectual property, especially right now when there is so much sensitivity,” said Chris Jacquemin, head of digital strategy at talent agency WME. 

When Lourd called Altman, the power agent said it sounded like the tool had been trained on Johansson’s voice, according to people familiar with the interaction. He asked for an apology and for the voice to be removed. OpenAI has said the Sky voice was not trained on Johansson. 

Days went by without a resolution. Johansson assembled a legal team that included John Berlinski, who often represents talent in contract disputes, and worked with her three years ago on a legal battle with Walt Disney over her salary in the movie “Black Widow” that the two sides eventually settled. 

On May 15, they sent a letter asking that OpenAI stop using the voice and offer transparency about its origins. The tech company, through lawyers, offered the name of the voice-over casting directors that they’d worked with, but not the name of the actor hired, the people said.

Johansson’s camp sent a second letter on May 19 and the actress went public with the matter the next day. “In a time when we are all grappling with deepfakes and the protection of our own likeness, our own work, our own identities, I believe these are questions that deserve absolute clarity,” Johansson said in a statement at the time.

Altman’s defenders say the Sky voice has little in common with Johansson’s voice in “Her” beyond being female and described by outsiders as “flirty.” 

The Hollywood actors’ union, the Screen Actors Guild-American Federation of Television and Radio Artists, has been in communication with Johansson’s team since she went public with her accusations against Open AI and said it is championing federal legislation to protect performers from unauthorized digital replication of their voices and likenesses.

“It is encouraging that open AI clearly took the concerns she raised seriously,” said Duncan Crabtree-Ireland, the national executive director of SAG, calling the current landscape a “Wild West situation.”

In its most recent contract with major studios and streamers, SAG was able to score some victories in the use of AI and compensation for actors should their likeness be used, but it was not able to get the studios to agree to seek consent to use an actor’s voice or likeness for training AI.

“At the end of the day, these AI platforms won’t police themselves,” said Dan Neely, CEO of Vermillio a generative AI startup focused on authorized use of intellectual property and talent in AI.

—Jessica Toonkel and Tom Dotan contributed to this article.

Write to Sarah Krouse at sarah.krouse@wsj.com, Deepa Seetharaman at deepa.seetharaman@wsj.com and Joe Flint at Joe.Flint@wsj.com