During Hollywood’s dual strikes, actors were offered paid work recording emotional performances for an AI study. The project, run by Realeyes, gathered material intended to help Meta train virtual avatars and support the company’s work on understanding and expressing human emotions. For participants, the opportunity came with a difficult question: what exactly had they agreed their data could be used for?
A paid study with a wider reach
The study ran from July through September and recruited actors. One participant, identified as T, described arriving at a rented Hollywood studio in early September for a two-hour shoot. He performed in front of three cameras, responding to prompts that called for anger, sadness, or fear.
The job posting offered $150 per hour for at least two hours and described the work as research. It also said an individual’s likeness would not be used for commercial purposes. T understood that the recordings would not be made public, and learned about Meta’s involvement only after arriving at the studio.
That framing made the work appealing to actors with limited opportunities during the strike. T said the session felt like an acting workshop, and he saw it as a chance to earn money without crossing the picket line. The posting also said the project did not qualify as struck work.
But a research label did not resolve how the recorded material could be used later. Participants signed a data license agreement that, according to MIT Technology Review’s review, gave Realeyes broad rights over the data and works derived from it. The agreement covered formats and media existing at the time or developed in the future.
What the agreement left uncertain
The central concern was the gap between the job posting’s assurance about individual likenesses and the agreement’s expansive language. The source reports that participants appeared to authorize Realeyes, Meta, and other parties chosen by the companies to use their faces, expressions, and material derived from them, provided they did not reproduce individual likenesses.
That distinction may be hard to assess in practice. A person’s face might not be reproduced as a recognizable copy, yet recordings of facial movements and expressions could still help train systems to create more humanlike outputs. Actors interviewed for the story worried that their contribution could shape AI tools or avatars in ways they could not foresee when signing.
Emily Poler, a litigator who represents clients in media, technology, and intellectual property disputes, said that agreements can grant rights over uses or technologies that did not exist when the contract was signed. She raised the question of whether someone can knowingly waive rights to things that do not yet exist.
Realeyes vice president for growth and marketing Max Kalehoff said the project was research and that its timing alongside the strikes was an “unfortunate coincidence.” He described possible applications including digital media experiences, medical diagnoses, safety alertness detection, and robotic tools supporting disorders related to facial-expression recognition. Meta did not respond to multiple requests for comment.
Why actors’ data matters to AI
AI systems rely on data, and people often supply the work and material that make it useful. Human workers prepare and label examples so machines can interpret them. For systems that analyze or generate humanlike faces, voices, and behavior, recordings of people can provide that material.
The article describes a shift from collecting publicly available data toward building datasets with recruited and compensated participants. That approach can provide higher-quality material and address concerns about whether data was gathered ethically and with proper consent. But compensation and a signed form do not automatically make the future use clear to a participant.
For performers, the issue connects to publicity rights: control over the use of a person’s name and likeness. Generative AI can use photos and videos as training material to create realistic replicas. The concern is not limited to celebrities. Background actors have also faced uncertainty about digital scans made on set, including how long studios may retain rights to use them.
A labor question beyond the study
The project’s timing sharpened an existing debate. Actors were on strike amid concerns about AI, likeness rights, compensation, and informed consent. Study participants may have viewed their work as separate from entertainment production, but the data they provided could still contribute to systems that perform humanlike tasks.
Jessica, a New York–based background actor who asked to be identified by her first name, called the project “exploitative.” She worried that companies could use recorded performances to generate composite avatars that look human without reproducing any one participant’s recognizable face. In her view, that could affect opportunities for actors even when an AI output is not a direct copy of them.
The dispute is therefore about more than whether a recording appears in a film or advertisement. It is also about how broadly a company can define likeness, whether consent covers future uses, and whether the pay and terms reflect the value of data that can be reused. As AI systems develop, actors may find that a short paid session has implications beyond the immediate job—while the agreement they signed leaves those implications difficult to predict.