Meta created and let users create flirty chatbots and intimate images using the names and likenesses of Taylor Swift and other celebs without their permission
Meta had already moved away from a short-lived celebrity chatbot feature toward user-customized AI avatars, making this episode a sharper test of how much control the company can delegate over identity use. That earlier shift to customizable AI avatars expanded the relevance of its moderation and consent rules beyond officially licensed personalities.
The issue also sits alongside earlier safeguards added after viral fake Taylor Swift images and later criticism of Meta’s image-generation product from privacy advocates, CAA, and SAG-AFTRA. The recurring pattern is that generative-AI product reach is outpacing settled protections for public Instagram users’ likenesses.
First-order effects
Celebrities named in the report face unauthorized sexualized representations and conversational personas tied to their identities, while Meta must address the availability of those creations on its products.
Meta’s AI-avatar and image-generation policies face immediate scrutiny over whether they adequately distinguish consented identity use from user prompts involving real people.
Second-order effects
Talent representatives and rights holders are likely to press AI platforms for clearer permission, removal, and escalation processes, raising the operational burden of celebrity- and likeness-based features.
Rivals developing image or avatar tools have a concrete warning that permissive handling of recognizable people can turn product-safety gaps into reputational and rights-management liabilities, as Microsoft’s earlier Designer safeguards following fake Swift images illustrate.
Third-order effects
If repeated incidents persist, likeness governance could become a core product-design constraint: platforms may need consent-aware controls rather than relying primarily on downstream moderation.
The dividing line between licensed celebrity experiences and user-generated impersonation may become a more consequential competitive and policy boundary, though the corpus does not establish what rules platforms or regulators will ultimately adopt.
The trend: Generative-AI platforms are moving from novelty celebrity features toward a durable contest over consent, identity control, and responsibility for user-created synthetic media.
Meta created AI chatbots mimicking celebrities like Taylor Swift and Scarlett Johansson without permission. These bots made flirtatious, explicit advances on Facebook, Instagram & WhatsApp. Meta removed them but faces ethics, legal and safety concerns over misuse of AI and cele…
The premise of AI as a business model is that no matter who you are, at any walk of life, anything about you can be appropriated by the billionaires and simulated for their profit [embedded post]
Yup. And trying to adjudicate that with classifiers is ... not easy. — This has been a persistent problem with edge content. In addition to violating content, Meta also classifies content as “borderline” and “ambiguous.” The latter two categories vastly outweigh the first.
A general journalism thought I've been having since jumping to Reuters: Part of responsibly covering things (tech, politics, wars, etc) is not sanitizing the details. — It's important that people know that the images Meta AI was/is creating are more than “provocative.” — My e…
The problem is that there's no technical method to “shut off” nude generation for a LLM, other than a) moderate the text input to see if the user is trying to generate nudes or b) moderate the generated output. — Then that gets into the “what counts as nude?” debate which has b…
It does seem like a hard thing not to be aware of for months on end, yes. — I'm going to be curious if they do choose to shut off Meta AI's existing capacity to produce near-nude images in the next few days.
It is trivial to prompt engineer any LLM to behave like a celebrity. The difference here is that even though LLMs are RLHFed to be syncophantic, getting them to *flirt* and send sexy images is a very intentional product design decision that could have been caught during QA if it…
“In several weeks of Reuters testing to observe the bots' behavior, the avatars often insisted they were the real actors and artists. The bots routinely made sexual advances, often inviting a test user for meet-ups.”
The miracle technology currently propping up the US economy has applications as varied as: plagiarism cheatbots, encouraging actual suicide, and creating fake porn of underage actresses or impersonation celebrity chatbots. Things are going just great. — www.reuters.com/busines…
Lawyer who I spoke to here seemed to think this is a clear-cut case of “not fair use.” But no celebrities wanted to comment for the story... Obviously deepfake porn has been out there for awhile from randos. Not sure the boundaries of the Overton window for a player the size o…
The extent to which these companies assume by default that they are unimpeachable and can do this stuff without any fear of meaningful consequences is reaching truly impressive heights. — Here is a Meta product lead just casually creating a sex bot out of the likeness of the mo…
I am really fucking tired of women's personhood, privacy, and agency being obliterated by tech companies that monetize our sexuality without our consent. — They shrug when we report rape threats and gendered harassment. Why? Because they're secretly encouraging this behavior.…
Many of the bots were requested by users, but a Meta gen AI product lead herself ordered up and shared chatbots impersonating Taylor Swift. Meta has deleted many of the bots and acknowledged that the intimate images Meta's systems produced violated the company's own policies.
Meta created celebrity romance chatbot replicas of Taylor Swift, Scarlett Johansson and Selena Gomez. Reuters found these “parody” bots would generate hyperrealistic images of the women in lingerie with their legs spread and invite users to the women's homes. [embedded post]