What Is “Personality Theft” in AI

In March 2026, Rolling Stone reported on a growing number of creators discovering AI-generated versions of themselves being used in ads and sponsored content they had never agreed to. Fashion and lifestyle creator Emy Brookins was among them, finding a video pulled from her TikTok with her speech and mouth movements altered to promote a product. She described it as personality theft. What was being taken wasn’t just her image, but the voice and presence she had spent years building with her audience.ith her audience.

That framing gets at something identity theft and likeness cloning don’t fully capture. What was used without permission wasn’t a credential or a single photograph. It was the trust built through consistent, recognizable creative presence, and someone else was putting it to work commercially.

This is the problem now being called personality theft. Unlike most of what we’ve covered in this space before, it has no legal name, no obvious detection method, and no clear framework for addressing it.

What Makes Personality Theft Different From Identity Theft

When most people think about AI identity violations, they think about faces and voices. A deepfake video that puts words in someone’s mouth. A voice clone used in a fake ad. These are real problems, and we’ve written about them in detail. But they operate on the observable layer of identity, things you can see or hear and, at least in principle, trace back to a specific recording or image.

Personality sits somewhere else entirely. By personality, we mean the accumulated layer of how someone expresses themselves: their tone, reasoning style, the rhythm of their sentences, their instinct for humor, the way they consistently show up in their work. None of those things live in a single file. They emerge across years of output, and that accumulation is exactly what makes them trainable by AI systems.

A creator’s credibility with their audience is built on that layer, not on any single piece of content. It’s what makes their recommendations trusted, their opinions worth reading, their endorsements commercially valuable. When it gets replicated without consent, something more than a file gets taken.

Why This Kind of Replication Is Harder to Detect

The difference comes down to what’s actually being copied. Likeness cloning needs a source file — a photo, a recording, a clip. Personality replication works from a pattern built across everything someone has published. AI models trained on a person’s public output absorb the decisions behind the work: the phrasing choices, the structural instincts, how they approach a subject. That gets encoded as a set of tendencies, and new content produced from that model can feel like a specific person without reproducing a single line they actually wrote or recorded.

Stanford University researchers put a number to how accurate this has become. They built AI agents from two-hour interviews with 1,052 people and found the agents matched their human counterparts with 85% accuracy across personality assessments and decision-making exercises. The researchers noted the same approach could model anyone with enough public output, without their knowledge. Consensual “digital twin” platforms already offer this commercially. The consent layer is a feature of those products, not the underlying technology, which means anyone can run the same process on a creator’s public archive without them ever knowing.

That’s why creators sit in the most exposed position. Their style is what makes them commercially distinct, and years of consistent public content make that style learnable. Independent creators in particular rarely have the legal resources to do anything about it when it happens.

The output doesn’t need to be identical to feel like someone. That gap between production and perception is where personality theft does most of its work.

What Personality Theft Actually Costs

The financial side, diverted ad revenue, stolen streams, unauthorized monetization, is the most visible and documented piece. Platforms are starting to respond: Spotify launched Artist Profile Protection in March 2026, letting creators review releases before they go live. But those tools address the distribution layer. What actually hits hardest rarely shows up in a payout report.

The harder question is what happens to the trust a creator has built when a replica of their presence is circulating somewhere they can’t see, attached to things they didn’t say, reaching people who have no reason to doubt it. That breaks down across four things that are genuinely difficult to recover.

1. Their credibility

Followers trust a creator’s recommendations because they’ve watched that person’s judgment play out across dozens of posts, videos, or episodes. A model that approximates their voice can issue those recommendations without the creator ever making them. Once that trust is spent on something they didn’t endorse, there’s no clean way to reclaim it.

2. Their personality

The specific mannerisms, instincts, and style that built an audience aren’t just characteristics. They’re the product. An AI trained on that output can generate new content in the same register — returning to the same concerns, landing humor the same way, framing ideas with the same cadence — and the creator has no say in what it produces or where it goes.

3. Their relationship with their audience

Audiences build parasocial relationships with creators through repeated exposure. They know their humor, their opinions, how they respond to things. When a replica approximates that presence well enough, some of that relationship transfers. The audience may never realize anything changed. The creator loses engagement they’ll never be able to account for.

4. Their commercial value

Brands pay creators because of that accumulated trust. A personality model that captures it can be used to simulate endorsements, redirect audiences, or build competing platforms on borrowed credibility. None of that requires the creator’s name on it. The mechanism is the same whether or not the impersonation is explicit.

For professionals outside the creator space, the same points apply in different form. A researcher whose reasoning is publicly documented across years of writing has a trainable personality. A model built on that output could attach their characteristic framing to conclusions they never reached, with no obvious fingerprint left behind.

Why Existing Laws Don’t Cover This

Understanding why personality theft is so hard to address legally starts with how intellectual property law was built. It was designed to protect specific things people made, not the patterns behind how they made them.

Copyright is the most obvious place to look. It protects specific works of expression, not the style or instincts behind them. A model trained on an artist’s catalog can absorb their creative patterns without reproducing anything existing doctrine would recognize as protected. The March 2026 lawsuit by independent musicians against Google raises a related question about training data consent, but that’s a different problem from stylistic replication, and a ruling on one doesn’t resolve the other.

The right of publicity gets closer, but its reach into expressive identity is largely untested in court. A voice clone in an ad falls within it. An AI-generated track that sounds like a specific artist but carries a different name sits in much less settled territory. Trademark doesn’t fill the gap either, since it protects brand signals, not how someone creates.

The expressive layer, where personality actually lives, sits outside every framework currently on the books.

The Shift From Content to Presence

The legal gaps explain what can’t be done right now. But personality theft also surfaces something broader about how identity functions in digital spaces.

Most frameworks for protecting identity, including most of what we’ve covered in this space from digital likeness to training data consent, treat identity as fixed and observable. A face. A name. A recorded voice. Something that can be verified against a source. What personality theft points to is a layer those systems weren’t built to reach: the ongoing presence a creator builds with their audience through consistent output over time.

That presence is now replicable. A model trained on a creator’s public output can generate new content that fits their characteristic patterns and feels consistent with their voice. The output isn’t them, but it registers as connected to them by people who engage with it. In a creator economy where presence is what audiences trust, that distinction carries real weight.

Until protection frameworks catch up to that reality, the people most affected are left managing a risk the current system was never designed to see.

FAQs