Meta Built an Opt-Out Button for Your Face. It Lasted Three Days.

On July 7, Meta announced Muse Image, a new AI image generator built into Instagram. Buried inside it was a feature that let any user type an @-mention of a public account and generate images referencing that person's photos. No notification. No request. The account holder found out the same way everyone else did: by seeing what strangers had made of their face.

The default setting was opt-out, as Variety reported. If you had a public Instagram account and didn't want to become raw material for someone else's AI image prompt, the burden was on you to go find the toggle and switch it off. Meta had, in effect, decided that your likeness was available for remixing unless you filed the paperwork to say otherwise. Complex's writeup on the rollout put it plainly: users over 18 with public accounts had to actively disable the feature to keep their photos out of it.

By July 10, the feature was gone. Meta's blog post, as TechCrunch documented, called it a feature that "missed the mark" and said it was "no longer available." Three days from launch to retreat, which by the standards of this industry counts as fast learning.

Except it isn't learning. It's a script, and we've seen the script before.

A company builds an AI capability that touches other people's data, images, or likeness without asking first. The design choice to make it opt-out rather than opt-in isn't an oversight; it's a bet that most people won't notice until it's already shipped, and that the ones who do notice will be a minority the product can absorb. Sometimes the bet pays off quietly, and the feature just becomes the water everyone swims in. Sometimes it doesn't, and there's a blog post about "missing the mark" a few days later. Either way, the next version of the same idea ships from someone else within months, with the same default.

What's notable about Muse Image isn't that it happened. It's how little friction there was on the way to shipping it. Somewhere inside Meta, a team scoped a feature that generates images of identifiable people from a database of billions of public photos, and the question "should this default to on or off" apparently resolved itself in favor of on. Not because anyone thinks strangers rendering your face without permission is fine in the abstract. Because opt-out ships faster, tests better in the aggregate engagement numbers, and only becomes a problem if enough people complain loudly enough, fast enough, before the next quarterly roadmap locks in.

This is the same company that spent this spring sending AI personas dressed as teenagers into competitors' chatbots to probe how they handle young users, a story we covered here in June. Different team, different product, same institutional instinct: treat the people on the other side of the interaction as inputs to be tested against, not parties who get a say. Consent, when it shows up at all, arrives as a settings-menu afterthought rather than a design constraint. It's the same instinct we wrote about when "AI-powered" started working against products instead of for them: the label alone doesn't earn trust, and neither does an apology issued only after the feature is already live in front of millions of people.

None of this required malice. It's what happens by default when the incentive structure rewards shipping speed and engagement metrics over asking permission first, and when "we'll fix it if people complain" is a viable strategy because complaints are cheap to absorb and features are expensive to build twice. The backlash cycle isn't a check on this behavior. It's a cost of doing business, priced in from the start, the same way a retailer prices in some amount of shoplifting rather than staffing every aisle.

The tell is in the language companies reach for when the retreat comes. "Missed the mark" implies aim. It implies the target was somewhere reasonable and the shot simply went wide, an unlucky miscalibration rather than a decision. Nobody says "we chose the setting that let this happen and we'd make the same choice again on the next feature unless the backlash gets loud enough." That's the sentence that would actually explain what happened, and it's the one you'll never see in a corporate blog post.

Instagram will ship another AI feature that touches other people's data before the year is out. The question worth asking isn't whether the backlash will come. It's whether the default will be different next time, or whether "missed the mark" is just the phrase they've decided to keep on file for whenever the same choice gets made again.