SAN FRANCISCO – Three days was all it took. Meta launched a new AI image feature on Instagram on July 7, let millions of users experiment with it, and pulled it on July 10 after creators and talent agencies raised urgent concerns about the generation of non-consensual imagery. The account owners whose photos fed the feature were never notified. There was no opt-out.
The feature was called Muse Image. It worked by letting any user type a text prompt and include an @ mention of any public Instagram account. Meta’s AI would then use that account’s publicly posted photos as a visual reference and generate an image from the prompt. A person whose photos were being referenced would receive no notification. There was no mechanism to block one’s likeness from being used, no way to opt out before the fact, no alert after it.
Meta has not disclosed how many users accessed Muse Image in those three days, how many accounts were referenced without the knowledge of their owners, or what kinds of images were generated.
What removed it was not an internal safety flag. The Creative Artists Agency – one of Hollywood’s largest talent firms, representing thousands of actors, musicians and entertainers – raised the alarm publicly, alongside a wider wave of criticism focused on the risk of non-consensual intimate imagery, known as NCII. NCII – synthetic content generated to depict real people in sexual or degrading situations without their consent – has become one of the most severe documented harms of generative AI deployment. Several US states and the United Kingdom have moved in recent years to criminalise its creation and distribution.
Meta’s statement after pulling the feature read: “Our intent was to provide a useful creative tool and to give people control over whether their public content could be referenced in this way. We’ve heard the feedback that this feature missed the mark, so it’s no longer available.”
The phrase “give people control” reveals the product’s central failure. That control – the ability for account owners to opt out of being referenced in Muse prompts – was not available when the feature launched. It was something Meta apparently planned to offer, not something it built before opening the tool to the public. Creators who discovered their faces had been synthesised into someone else’s AI content during those three days had no recourse at the time.
The episode arrives during a period of unusual legal pressure on how platforms handle AI image generation. Apple’s suit against OpenAI over alleged trade-secret theft marks one front of legal challenge over AI’s use of identifiable content. The consent questions raised by Muse are different in kind – generation rather than training – but share the same underlying assumption: that publicly available content is freely usable in any technical context, regardless of what its creator intended.
Meta has shipped AI features across Facebook, Instagram and WhatsApp in rapid succession over the past eighteen months. AI-generated summaries, AI responses in comment threads, AI-generated image filters – each has arrived before the company’s stated safety policies have been fully mapped to the new capability. Muse Image extended that pattern into territory involving identity and physical appearance, which is precisely where AI-generated content causes the most irreversible harm.
Meta’s own policies prohibit the sharing of NCII across its platforms. Whether Muse Image could generate content that violated those policies is a question the company has not answered publicly. TechCrunch reported on the feature’s launch and withdrawal, noting that Meta offered no account of what its internal safety review evaluated before launch.
For creators and entertainers, the episode confirms a pattern many have described since generative AI tools became mainstream: platforms extract first and seek permission when challenged. CAA’s intervention produced a result in three days. Most account owners whose photos were used as references in Muse Image prompts would never know. And most lack CAA’s institutional weight to compel any response.
The growing volume of consequential decisions about AI and identifiable human content spans sectors far beyond social platforms. The same year that philanthropic AI investments are being directed toward cancer research, generative tools embedded in consumer platforms are extracting human likenesses without consent frameworks in place. The governing logic in each context is different; the pace of deployment is not.
Meta has not said whether Muse Image will return in a modified form, what modification would constitute a satisfactory consent mechanism, or when the company plans to address the gap its own statement identified. The apology offered when the feature was pulled committed to nothing specific. What remains open – whether account owners will have a meaningful voice in how their likeness is used by AI tools built into the platforms they publish on – is exactly the question the AI industry has not resolved.
