Meta has killed an Instagram feature that let users generate AI-modified images of any public account by @-mentioning it, pulling the tool on Friday just days after rolling it out. The feature shipped earlier the same week as part of Muse Image, a new AI image generator built by Meta Superintelligence Labs. Users whose photos were referenced received no notification when their likeness was fed into someone else's prompt.
The company said in a Friday blog post that the feature was no longer available and that it had 'heard the feedback' it 'missed the mark.' Puck News founding partner Dylan Byers was first to report the reversal, noting the decision came 'amid scrutiny from users and talent agencies, including CAA.'
The launch-to-kill cycle ran roughly four days. Muse Image itself remains live; only the @-mention referencing capability, which effectively turned every public Instagram profile into training data on demand for any other user's prompt, has been withdrawn. Meta had positioned the tool as a creative aid, with an opt-out for users who did not want their public content referenced.
Key facts
- 01Meta removed an Instagram AI feature on Friday, days after launching it earlier the same week as part of Muse Image.
- 02The feature let users generate AI images by @-mentioning any public Instagram account, with no notification to the referenced user.
- 03Meta Superintelligence Labs, the company's dedicated AI unit, built Muse Image.
- 04Talent agency CAA was among the parties pressuring Meta over the feature, per Puck News founding partner Dylan Byers.
- 05Meta said the feature 'missed the mark' in a Friday blog post announcing the reversal.
That opt-out design was the immediate flashpoint. Making the feature default-on for every public account, with no in-the-moment alert when someone's photos were pulled into a generation, inverted the burden: subjects had to know the feature existed and find the setting to disable it. Coverage of how to turn the feature off circulated widely in the days between launch and withdrawal.
“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.”— Meta, company blog post
CAA's involvement points to the specific commercial exposure that likely accelerated the decision. The agency represents actors, musicians, and athletes whose public Instagram accounts are, by professional necessity, public — and whose likenesses carry contractual protections that a consumer AI tool referencing them by @-mention would immediately complicate. A talent agency raising the issue is a faster signal to legal than user complaints alone.
The pattern of AI features being deployed on social platforms and then abused to generate non-consensual imagery, particularly of women, is now well-documented across the industry. Guardrails announced at launch have repeatedly proven insufficient once features reach scale. Meta's decision to pull the tool within days rather than iterate on filters suggests the company concluded no filter set would survive contact with the user base.
Muse Image and Meta Superintelligence Labs remain central to Meta's consumer AI push, and this reversal does not appear to affect other Muse Image capabilities. The company has not indicated whether a redesigned version of the referencing feature, with consent gating or opt-in defaults, will return.
For Meta, the cost of the episode is not the feature itself but the signal it sends about internal review. Shipping a tool that lets any user generate AI imagery referencing any public account, without notification to the referenced party, is the kind of design choice that a functioning trust-and-safety review is supposed to catch before launch, not four days after. That the fix arrived only after CAA weighed in is the part product leadership will have to answer for internally.
The broader read for the AI-on-social-platforms category is that consent architecture is now the load-bearing question, not model quality. Muse Image is presumably capable; the referencing feature was pulled on governance grounds, not output grounds. Platforms shipping generative tools into feeds full of real people's faces will increasingly find that the deployment surface, not the model, is where the product lives or dies.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
