Meta pulls Instagram's Muse Image @-mention feature three days after launch

The problem

On July 7, 2026, Meta introduced Muse Image, the first image-generation model from its Meta Superintelligence Labs unit, and promoted a way to use it inside Meta AI: users could generate images by @-mentioning public Instagram accounts that they wanted to reference, pulling publicly available photos from those accounts as references [source]. Adults with public Instagram accounts were enrolled by default and had to open their settings to opt out, and the feature was not designed to alert a user if their photos were used in this way ([source]).

The design prompted immediate backlash over non-consensual use of people’s likenesses [source]. CAA said that “No one’s name, image, likeness, voice, or creative work should be used by any third party, including AI models, without clear, documented consent” ([source]). SAG-AFTRA said that “With the dangers of nonconsensual digital replicas well known to all, a feature that encouraged that behavior is unwise” [source].

What was built

The @-mention capability of Muse Image was built by Meta Superintelligence Labs and surfaced through Meta AI and Instagram [source]. Only that feature is at issue here: the Muse Image model itself and its access via other Meta surfaces were not the subject of the removal ([source]).

The outcome

On July 10, 2026, three days after launch, Meta reversed course and removed the feature [source]. In its own Newsroom post, updated July 10 at 3:45 PM PT, Meta stated that “We’ve heard the feedback that this feature missed the mark, so it’s no longer available” ([source]). After the removal, CAA welcomed the decision, saying “We commend Meta for its swift decision to remove the Muse Image feature” [source].

How the sources hold up

The strongest source here is Meta’s own Newsroom post on about.fb.com, a first-party (Tier 1) against-interest admission on Meta’s own domain, located and archived, so the “missed the mark” statement is directly retrieved rather than merely press-attributed [source]. The weakest load-bearing sources are the three independent newsrooms: TechCrunch, TechRepublic and Music Business Worldwide are strong secondary (Tier 2) reporting, not primary documents, so the CAA and SAG-AFTRA statements rest on press attribution rather than on each organization’s own filing ([source]).

Related case files

• Slack changed its AI privacy principles after backlash over an opt-out-by-default use of user messages, the same enroll-by-default-then-retreat pattern Meta ran here.
• Mozilla rewrote Firefox’s terms of use after backlash to state it does not use your data to train AI, another consent-and-likeness reversal driven by public objection rather than a regulator.
• X suspended EU user-data training for Grok after the Irish DPC took High Court action, where the trigger was a regulator rather than talent-agency pressure, but the retreat was the same.