Videos that appear to show familiar celebrities endorsing products have become a tool for online scams. Tom Hanks, Gayle King and MrBeast each warned that their likenesses had been used without permission to promote offers they were not associated with.
Three warnings, three unauthorized promotions
Hanks posted on Instagram after an AI-generated version of him appeared in an ad for a dental plan. He said he had nothing to do with it. The following day, CBS anchor Gayle King flagged a video using her likeness to promote a weight-loss product, saying she had never heard of or used the product.
MrBeast described another example on X: a modified video that appeared to show him promoting a fraudulent iPhone 15 giveaway. The clip manipulated both his speech and lip movements. He said many people were seeing the scam ad and questioned whether social media platforms were ready for the spread of AI deepfakes.
The promotions use recognizable people to make dubious offers seem credible. A viewer may see a familiar face and voice and assume the person has approved the product or giveaway. In these cases, the public warnings from the people impersonated made clear that the apparent endorsements were false.
How the videos may be made
The original Hanks video was not available for review. Based on examples shared by King and MrBeast, the likely method was to start with existing celebrity footage, alter the lip movements and pair it with AI-generated voice clones.
Those voice clones may be trained on vocal samples from publicly available work. The result can make a familiar person appear to say something new, even though the person had no involvement in the advertisement. The technique combines recognizable footage with synthetic speech to create a misleading endorsement.
That distinction matters: the video may resemble a real recording while its message is fabricated. A polished clip alone does not establish that a celebrity supports a product, giveaway or claim.
Impersonation raises wider questions
The scams arrive amid broader debate about the use of AI in entertainment and media. Concerns about AI were a significant point of contention during the Writers Guild of America strike. SAG-AFTRA, which represents Hollywood actors, has also expressed concern that AI could create digital replicas of performers without proper approval or compensation.
Voice cloning has also affected people beyond those targeted in product ads. Zelda Williams, Robin Williams’ daughter, made headlines after objecting to people cloning her late father’s voice without permission. Together, these examples raise questions about who can use a person’s likeness or voice, and under what circumstances.
For audiences, the harm extends beyond any single promotion. When synthetic video can make someone appear to endorse an offer, it can become harder to judge whether digital media is authentic. Repeated impersonations may also weaken confidence in real recordings and in the identity of the person speaking.
Why detection and moderation are difficult
Watermarks and metadata could help identify AI-generated material and track where it came from. Google and OpenAI have plans to use these approaches, but watermarks have historically been easy to defeat. Open-source AI tools that do not add watermarks are also available, which limits how much a label can solve on its own.
Regulation presents a different challenge. Restricting AI software could affect legitimate researchers while leaving tools available to people who intend to use them for fraud. The source of a video and the context in which it is shared may therefore remain important even when technical labels are present.
Social media networks may need to respond quickly when users flag suspicious content. The celebrity warnings show how impersonated people can help alert viewers, but effective moderation also depends on platforms acting on reports. For now, these cases leave audiences with a practical reason to check whether an apparent endorsement has been confirmed by the person whose face and voice appear in the video.