Meta launched Muse Image last Tuesday. It was the first image model out of Meta Superintelligence Labs, built into the Meta AI chatbot, and it came with a feature, launched in the US, that let anyone generate AI images by tagging a public Instagram account. By Friday evening on the same week, that feature was gone. Meta’s statement said it had “missed the mark”, which is about as close to a public apology as a platform of that size ever gets.
The speed of the reversal is what most of the coverage focused on. I think the more useful story for marketers is what happened between Tuesday and Friday, because Meta did not fold straight away with Muse Image. When Creative Artists Agency, the Hollywood talent agency known as CAA, published its objection on Wednesday, Meta’s first response was to defend the design. The company pointed out that private accounts and under-18s were automatically excluded, and that adults with public accounts could opt out in a couple of clicks. Within roughly 48 hours of saying that, the feature was dead.
I have sat in enough launch reviews over the years to recognise that sequence. A team ships something it believes in, the objection arrives, the first instinct is to explain the safeguards, and only then does someone realise the safeguards were never the issue. The issue was the assumption underneath them.
Permission is the relationship, not a setting inside it
In my marketing lectures, I teach customer relationship management as the process of building and maintaining profitable customer relationships by delivering superior value and satisfaction. Trust and credibility sit at the centre of that definition. When customers trust a brand, they feel confident in their decisions and view the brand as a reliable partner. Every framework in that lecture assumes one thing before any of it works, which is that the customer agreed to be in the relationship in the first place.
Well, Muse Image skipped that step. Public accounts belonging to adults were included by default, with no notification sent to the account holders. Meta treated the absence of a private setting as the presence of consent, and those are not the same thing. Posting a photo publicly means you are happy for people to see it. It does not mean you have agreed for strangers to feed it into an image generator and remix your likeness.
The distinction matters commercially, not just ethically. Customer equity is built on the accumulated lifetime value of people who choose to stay. A feature that makes millions of users feel like raw material rather than customers spends down that equity for a product benefit most of them never asked for.
The default is the message
Meta’s defence, that opting out took only a couple of clicks, answers the wrong question. Nobody was disputing how easy the switch was to find. The objection was about which way the switch pointed when nobody touched it.
Defaults are not a settings decision. They are a statement of what the company assumed about you. Opt-in says we will wait until you are ready. Opt-out says we have already decided on your behalf, and the burden of undoing that decision is yours. SAG-AFTRA, the union representing Hollywood actors and performers, made exactly this point when it told members that anything short of a clear and conspicuous opt-in was unacceptable for uses of this kind. When the feature was withdrawn, the union called the discontinuance “the responsible thing to do”.
Marketers have known this trade-off for decades. Every email programme, every loyalty scheme, every data consent flow forces the same choice between short-term scale and long-term trust. Opt-out always wins the volume metric on day one. It just tends to lose everything else.
The precedent was sitting right there
This was not a novel mistake for Muse Image. OpenAI ran the same play with Sora last year, launching with an opt-out approach to likenesses and IP, absorbing condemnation from CAA and the Motion Picture Association, pivoting to opt-in under pressure, and eventually shutting the product down. The entire case study existed, publicly documented, before Meta’s launch review would have happened.
That is what makes this a governance failure rather than a design one. And it helps to understand the pressure Meta was under. Muse Image is only the second major release from Meta Superintelligence Labs, the expensively assembled unit whose first model replaced the Llama family entirely, and until this launch Meta’s image tools ran on third-party models from the likes of Midjourney and Black Forest Labs.
Meta’s own benchmarks place Muse Image behind OpenAI’s GPT Image 2 overall. So the company could not win on raw model quality, and its differentiator was the one thing OpenAI and Google cannot touch, which is your social graph. The @-mention feature was not a careless add-on. It was the moat. The feature carrying the biggest consent risk was also the feature carrying the entire strategic rationale, and that is exactly the condition under which launch governance gets steamrolled. Nobody in that room wanted to kill the only thing that made the product different.
I wrote recently about Samsung’s AI governance arc, and the contrast is instructive. Governance that works catches the consent question before launch, because someone in the room is paid to ask it. Governance that does not work discovers the question on day two, from a talent agency’s press statement, and then spends the rest of the week deciding whether to defend or retreat.
Meta kept Muse Image alive in the Meta AI app and on WhatsApp, so this was a targeted amputation rather than a product death. The model survived. What did not survive was the assumption that public content is automatically available content, and I suspect that assumption will now be tested in every AI launch review across the industry, which might be the most useful thing to come out of the whole episode.
One more thing if you are reading this in Asia
The generation feature never launched here. The permission did. Meta switched the setting on by default for public accounts outside the US as well, so your account was sitting in the pool before the tool ever reached the region. On the mobile app, the control sits under Settings and activity, in the How Others Can Interact With You section, listed as Sharing and Reuse, and Tom’s Guide has a step-by-step walkthrough if you want to switch it off. If you cannot see it, that is not you. The control is still rolling out beyond the US, it does not appear on the desktop site, and it may not show on private accounts, which were excluded from the feature anyway.
I should say I could not verify this on my own phone. My account is private, so the setting does not show for me at all, which is consistent with private accounts being left out of the feature. If you are on a public account in this part of the world, I would genuinely like to know whether the setting has reached you yet, so do tell me in the comments.
The default arrived everywhere before the off switch did, which rather makes the point of this piece for me. Two things are worth knowing once it does appear. Opting out only protects what you post from that point on, so the earlier you switch it off, the less of your content sits in the reusable pool if and when the feature returns. And the same setting controls Instagram’s collaboration features like Remixes and templates, so creators who rely on public reach are being asked to trade a little distribution for a lot of control. That trade-off is the whole story of this piece in a single settings screen.
The lesson costs nothing to learn second-hand. Permission is not friction to be engineered around. It is the first thing a customer ever gives you, and the way you ask for it tells them everything about what kind of relationship they are in.
