By: Scott Harris
Why AI Training Data May Be the Next Frontier of Digital Persona Rights
So far in this series I’ve written about deepfakes, digital footprints, and what happens to your identity after you die. All of them lead back to a question that’s quietly shaping the whole debate: who owns the data that teaches AI to imitate people? The answer may decide who controls identity in a world where machines can learn to act like us.
The lawsuits making headlines
Over the last few years, a wave of high-profile lawsuits has gone after how AI systems are trained. Media companies, artists, photographers, and authors argue that their work was used without permission to build commercial AI products. A few of the big ones:
- The New York Times lawsuit against OpenAI and Microsoft, alleging that copyrighted news articles were used to train AI systems without authorization
- The lawsuit from Getty Images against Stability AI, claiming that millions of copyrighted photographs were used to train image-generation models
- Multiple lawsuits from authors, including claims organized through the Authors Guild, arguing that books were used in AI training datasets without consent
These are copyright cases, and the argument is fairly simple: if you own the copyright to a photo, article, song, or book, using it to train AI may infringe on that right. But copyright protects creative works. It doesn’t protect people, and that distinction may turn out to be one of the most important legal questions of the AI era.
Copyright protects content, not people
Creators have real protections. A photographer owns the photo, a musician owns their recordings and compositions, and a writer owns their work. When AI is trained on that material, disputes follow. One example was the viral AI-generated track “Heart on My Sleeve,” which used imitated voices of Drake and The Weeknd. After a public and legal backlash, the song was removed from streaming platforms within days, though it’s estimated to have racked up tens of millions of streams before the takedowns, and reuploads kept circulating.
Most people aren’t professional creators, though. We share photos, videos, voice clips, comments, and stories online without ever thinking about intellectual property. We’re just living our lives and staying in touch. And your face, your voice, your mannerisms, and your personality generally aren’t protected by copyright at all. They belong to you, but the law doesn’t treat them the way it treats a song or a photograph. Copyright can protect a photo of you without protecting the identity in it. AI is exposing that gap in ways lawmakers never anticipated.
AI is learning from all of us
A common assumption is that AI mostly learns from professional content like books, news articles, and research papers. In reality, much of the internet is made by ordinary people: social media photos, YouTube videos, podcasts, voice recordings, blog posts, comments, reviews, and livestreams. Together, that’s one of the largest records of human behavior ever assembled, and AI doesn’t necessarily distinguish between a celebrity’s content and yours. Both can end up teaching machines how people talk, show emotion, solve problems, and interact.
In other words, AI is learning from society itself. In January 2025, for example, LinkedIn Premium users filed a proposed class action alleging that a setting allowing their data to be used for AI training had been switched on by default without clear disclosure, a claim LinkedIn said had no merit. Disputes like that highlight a real concern: the content and profile data we share may get used in ways we never expected when we signed up, especially when terms of service and data policies change over time.
Whose identity is it?
This is where Digital Persona Rights come in. If an AI system can produce a voice that sounds like yours, a character that looks like you, or writing that reads like yours, the issue isn’t really copyright anymore. The question becomes: who owns the rights to the identity being replicated?
Celebrities sometimes have answers through publicity rights, endorsement deals, and contracts. For everyone else, the answer is far less clear, and most people have little protection if their likeness, voice, or behavior ends up indirectly built into an AI system.
The scale makes it bigger still. Large models are trained on hundreds of billions of images and trillions of words gathered from across the internet, including decades of posts, photos, and videos from people who never imagined they’d help train AI. Any one person is a tiny fraction of the data, but billions of us together are helping build systems that can replicate human speech, creativity, appearance, and behavior, mostly without knowing it. AI gets its value from human-generated data, and the humans behind that data usually have no idea how it’s being used.
Public isn’t the same as personal
Some argue that anything public online should be fair game for training. That holds up better for public information or published creative work. Identity feels different to me. Posting a photo doesn’t mean you agreed to have your face become part of a global dataset for generating synthetic people. Sharing a video doesn’t mean you consented to having your voice analyzed and reproduced. Leaving comments doesn’t mean you signed up to have your writing style modeled.
People draw these lines all the time. Plenty of parents object to strangers photographing their kids at a ballgame while happily posting pictures of those same kids online. That isn’t hypocrisy. It’s the difference between sharing something with people you trust and having it taken by someone you don’t. (I’ll come back to that in the next post.) The gap between sharing content and training AI on it is much bigger than most people realize, and it’s going to get harder to ignore.
Value without compensation
There’s an economic side to this as well. Many of today’s most valuable AI systems were built on information created by millions of people who rarely receive compensation, credit, or even notice. If AI companies profit from systems trained on our data, should we have any stake in the value created? Some researchers have proposed data licensing, data dividends, or consent-based training. Others argue that public information should stay freely available to encourage innovation. Either way, it shows that Digital Persona Rights will eventually touch economics as well as privacy and identity.
Musicians are already pushing back. In June 2026, the American Federation of Musicians sued Warner Music Group and Universal Music Group, alleging that the labels allowed AI companies to train on union musicians’ recordings without the performers’ authorization or compensation.
Where MiMir stands
Since I’m asking these questions of other companies, I should answer them for my own. MiMir does not use our customers’ data or stories to train any outside AI. We don’t use external AI services to access or even optimize customer data. Our AI tools run locally, and the only training we do that involves customer data is to improve our own internal AI so it can better transcribe, translate, index, and organize your stories inside the MiMir app. Your data is never used to advance any AI model that exists outside the MiMir app and our data center.
I don’t think that should be unusual. I think it should be the norm.
A new category of rights
That’s why I believe the next phase of digital rights has to recognize identity as its own protected category. Digital Persona Rights could fill gaps copyright was never designed to cover, including:
- training datasets that contain personal likenesses
- AI-generated replicas of faces or voices
- behavioral and personality modeling
- unauthorized digital avatars
- consent requirements for identity-based AI systems
These rights wouldn’t replace copyright. They’d cover something copyright was never meant to govern: ownership and control of identity itself.
For decades, the law has focused on protecting content. AI can now learn from something far more personal: the face in a photo, the voice in a recording, the style of a post, and the patterns in years of online activity. The debate is shifting from who owns information to who owns the identity reflected in it. Whether future laws treat identity as property, as personal data, or as its own kind of human right, I’m convinced of one thing: the people who teach AI how to be human deserve a say in how that learning happens.
Join the conversation