The Setting Nobody Checked

On July 8, Meta launched Muse Image, the first AI image generator to come out of its new Meta Superintelligence Labs division. You could reach it through the Meta AI app, Instagram Stories, WhatsApp. It could generate, edit, and remix photos — and it could do that by pulling in images from public Instagram profiles, just by referencing the account.

Here's the part that matters: if your Instagram was public, you were already opted in. Nobody asked. Nobody sent a notification saying your photos could now be used as raw material for someone else's AI-generated image. The switch to turn it off existed, but it was buried in a settings menu most people have never opened, for a feature most people didn't know existed, to protect against a use case most people hadn't imagined yet.

That's not a technical detail. That's the entire story.

Less than three days later, Meta pulled the feature. SAG-AFTRA had told its members to opt out and protect their likeness. CAA raised copyright and consent concerns on behalf of its clients. Actors, creators, and privacy advocates said, in plain terms, that this was a tool for generating images of real people without their knowledge or agreement. Meta's statement was short: "We've heard the feedback that this feature missed the mark, so it's no longer available." Muse Video, the companion tool for AI video generation, is still live.

Three days. That's how long it took between launch and reversal. I keep coming back to that number, because it tells you two different things depending on how you look at it. From one angle, it's reassuring — the system worked, the backlash was fast, the correction was faster. From another angle, it's the opposite of reassuring: a feature with this scope shipped to the public without anyone internally raising a hand and asking whether "opted in by default" was a defensible choice for something that touches a person's face and likeness.

I think about this differently now than I would have five years ago, and it's not because I've become more anxious about technology. I've spent my career around it. I like what it can do. What's changed is that I have a son, and I think about the internet less as a tool I use and more as an environment he's going to grow up inside of — one that's already deciding, on his behalf, what counts as fair use of his image before he's old enough to have an opinion about it.

That's the piece that gets lost when a story like this gets filed under "tech company walks back feature." It sounds routine. Companies ship things, users complain, companies adjust — that's the news cycle working as designed. But the reason this deserves more attention than a routine correction is that the thing being tested wasn't a UI preference or a notification setting. It was whether your image — the actual photograph of your actual face — could be treated as public material the moment your account was set to public. That's a genuinely new question, and the default answer Meta chose, even briefly, was yes.

We've gotten used to talking about AI in terms of what it can produce — better images, faster video, more convincing text. Less attention goes to who gets to decide what feeds into it, and on what terms. Muse Image is a useful case study because it makes that question concrete. It wasn't asking "can AI generate a compelling image." It was asking "can a platform use your face as training or reference material for someone else's creative output, without telling you, unless you find your way to a setting you didn't know to look for." Those are different questions, and only one of them got the attention it deserved before launch.

I don't think the instinct to build this was malicious. I think it's something more familiar and, in a way, more concerning: a team moved fast, treated consent as a settings-menu problem instead of a design problem, and assumed that public meant available. That assumption used to be closer to true. A public Instagram photo meant other people could see it. It didn't used to mean a company could feed it into a generative model on your behalf, without asking, and let someone else remix your likeness into whatever they wanted to make. The definition of "public" is quietly expanding to include uses nobody agreed to when they flipped a privacy toggle years ago, and most people have no idea the ground has shifted under them.

This is why I don't think we can afford to treat stories like this as noise. AI gets introduced casually — a new feature, a new toggle, a new "creative tool" — and the pace of that introduction outstrips most people's ability to understand what they just agreed to, or didn't. When something moves that fast, it's tempting to let it pass, to assume the correction proves the system is self-regulating, to move on to the next headline. I'd push back on that instinct. A three-day turnaround is not the same as the problem being solved. It's evidence that the problem was obvious enough, and the backlash loud enough, that this particular instance got walked back. The next one might not attract a union statement or a Hollywood agency's attention. It might just quietly become normal.

None of this means the underlying idea — AI tools that can edit and remix images — is the problem. Plenty of that work is genuinely useful, and I'm not interested in writing the version of this piece that treats every new AI feature as a threat. The problem was never the innovation. It was the rollout: a default setting that favored engagement over consent, an opt-out that assumed most people wouldn't find it, and a company that only reconsidered after the backlash became too loud to manage.

If you have a public Instagram account — or any account tied to your image, your kid's images, your name — this is a reasonable moment to go check what you've agreed to. Not because Muse Image is still live. It isn't. But because the setting governing whether your content can be referenced by tools like it almost certainly still exists, defaulted to whatever the platform chose for you, waiting for the next feature that decides "public" means "fair game."

We're at a point where the tools are changing faster than most people's understanding of what they're agreeing to. That gap is where the real risk lives — not in the AI itself, but in how quietly the terms of consent can shift while everyone's attention is on the next demo.