The Rundown: AI clones split the creator economy

Aoden Teo, co-founder of Miso Labs, had previously posted that the company could “clone any voice with just 10 seconds of audio” and confirmed that another model was based on creator Salman Khan, adding “Pls don’t sue us.” Digiday reached out to both Mongeau and Miso Labs for comment, but did not receive a response. The video is still available at the time of writing.

Miso is a vivid example of a broader shift: generative AI tools can now mimic a creator’s look and sound from just a few seconds of content scraped from the internet. That puts creators and brands in a bind: either embrace AI and try to control them, or risk having lookalikes siphon off attention, deals, and reputation with few clear legal protections.

Why does this matter for creators and brands?

A creator builds their career and business off of their likeness, so anything that can convincingly mimic them holds immense power. For brands, choosing to partner with a creator involves a level of trust and, in some cases, a willingness to be less precious with their brand identity to allow for creative freedom and give creators the space to make content that will resonate with their audiences. 

AI clones (authorized or otherwise) can act as stand-ins for creators, and if they aren’t given the proper guardrails, can potentially go wildly off-script. This could be bad for both brand deals, the creator’s reputation, and even their bottom line — AI clones could cannibalize engagement and pull attention, and potentially revenue, away from the original creators. 

What’s changed?

Large language models, or LLMs, are rapidly evolving, moving from more simplistic, text-based generations into full-blown AI-generated audio and video in a short period of time. More companies are creating their own models and pulling data from every corner of the internet to train them and aid in generation.

“Creators need to understand that the biggest copier of them is Google Gemini,” said Gautam Goswami, CEO and chief AI officer at POP.STORE and ECHO-ME.AI. “Nano Banana [powered by Gemini 3] has literally sucked up the pictures of every public person on the internet and created a model based on that… if someone is trying to create somebody new, the chances that they will copy another creator are so high.”

Because of how these systems are trained, creators’ likenesses can be copied — sometimes unintentionally — even when no one sets out to clone a specific person, Goswami alleged. 

The advanced nature of these newer models means that they need smaller and smaller data inputs to create sophisticated clones. Miso Labs’ Teo, for instance, boasted that the company’s voice model only needs 10 seconds of audio to clone a voice. That combination of speed, ease, and light regulation leaves creators in a tough spot.

What about the law?

Creators hoping to protect their likenesses from being cloned with AI may have to trademark themselves, according to Frank Poe, attorney and founder of creator-focused firm, Poe Law. 

The examples Poe gave were celebrities attempting to trademark specific phrases (Taylor Swift filing for “Hey, it’s Taylor Swift” and “Hey, it’s Taylor” and Matthew McConaughey filing for “alright, alright, alright”) not their voices outright. McConaughey’s team also filed trademark applications for two different clips of him, to “create a clear perimeter around ownership with consent and attribution the norm in an AI world,” according to the Wall Street Journal. Taylor Swift also filed applications for a photo of her during her Eras Tour, according to the BBC.

This isn’t a failsafe, however. Not only does the internet make trademark law murkier, but the legal process is intensive. And creators may have to prove damages as well as that the company in question knowingly used their content to train its AI models without permission, according to Poe. 

“It’s very difficult to prove… it comes down to litigation. You have to do discovery, you have to attempt to subpoena, to uncork how this model was trained and where they got sources from,” Poe said. 

Pulling from YouTube videos could veer into copyright infringement, but that would still need to be proven in court, which is very time and resource intensive. 

“The question becomes, ‘is there a public policy to now allow them to farm these kinds of things’?” Poe said.

How are creators responding?

Some creators are choosing to work with generative AI companies to build their own clones, to varying degrees of success. Early this year, TikTok star Khaby Lame (who has nearly 162 million followers on the platform) inked a $975 million deal with Rich Sparkle Holdings, which licensed the rights to his AI-generated digital twin, which Vanity Fair reported would be used to sell products and fulfill brand deals in Lame’s stead. In the months since, the share price of Rich Sparkle Holdings have fallen by more than 90%, and Lame has remained quiet about the deal. 

TikToker Vicky “Whoa Vicky” Waldrip released an AI clone of herself last year that fans can call or message. But a Waldrip spokesperson told Digiday the clone has since been “discontinued.” Meanwhile, Bravo star and reality TV show producer Andy Cohen has his own AI avatar on the Peacock app. 

POP.STORE lets creators make AI versions of themselves, and Goswami told Digiday it’s a way for creators to own their AI likeness. The profiles built on the platform are not shared outside of its ecosystem. 

“When they delete their account, or when they delete the profile, it’s gone, gone. It’s not sitting in a shared database of any kind,” Goswami said. “It’s not like if they go and train their face on Meta’s AI. If they do, there’s no way of taking it back, because Meta’s AI will learn their face and their face will be imprinted in that AI’s memory forever.”

Napster is taking a similar route, according to communications manager Sean Morrison. In recent years, the company has shifted towards AI tools and SaaS products, launching a “real-time generative web and agentic AI video chat service” last May that lets creators build digital twins of themselves. Morrison says there’s more business-related use cases for the product, like a chief technology officer making a digital twin to set up one-on-one meetings with everyone at the company.

But crucially, if an individual deletes their digital twin with Napster, it’s deleted in its entirety — no digital echoes reverberating across the web. And the company is very careful in terms of rights ownership.

“The industry is kind of in a moment right now where it’s full speed ahead,” Morrison said. “We kind of take a different tack with it, where those consequences have to be front and center, and acknowledge them.”

But if creators want nothing to do with digital twins and AI clones, it’s unclear how much recourse they have.

What’s next?

Goswami suggested governmental regulation is the only way to curb the spread of unauthorized AI clones. 

“There’s no regulation… it’s not like in the Nordic countries, where you own your likeness,” he said, referencing Denmark’s push to fight deepfakes by giving people the copyright to their own features. 

Aside from attempting to copyright their voice and face, creators could also look to fold AI protections into their future contracts, whether that’s by splitting up their likeness rights or pushing for “no training” clauses. 

Expect more lines drawn in the creator space in the very near future. 

“Over the next one or two years, I think AI clones will become more visible, but still remain a specialist tool rather than a default part of creator marketing,” global influencer marketing strategist Aleksei Poliakov told Digiday over email. “They will make sense for multilingual campaigns, entertainment formats, digital avatars and certain brand experiments. But for most creator partnerships, there are still too many unresolved questions around audience trust, transparency, likeness rights, approval and control.”