What happened
Meta introduced Muse Image on July 7, 2026, as its first image generation model from Meta Superintelligence Labs. The company said the tool was available in Meta AI and could blend photos, use presets, support direct edits, and power creative experiences across Instagram and WhatsApp.
The backlash centered on a feature that let people generate images by @-mentioning public Instagram accounts they wanted to reference. According to Meta's own July 10 update to the launch post, the company heard feedback that the feature missed the mark and made it unavailable.
AP reported on July 11 that the feature had made publicly posted Instagram photos usable as references for AI-generated images, prompting privacy concerns and opt-out instructions spreading across social media. The Verge reported the same basic sequence and noted criticism from groups concerned about likeness misuse and high-risk design.
Get permission for the intended use
Check whose material was used, whether they agreed to that use, and whether the result implies an endorsement or performance that never happened. An AI label alone does not establish consent.
Someone may agree to AI caption cleanup, transcript summaries, or clip-title suggestions without agreeing to the reuse of their face or voice in a synthetic asset.
This is why opt-out controls can feel backwards in identity-heavy workflows. If a system creates something that looks like a person, speaks like a person, or borrows the social meaning of that person's public presence, the burden should not quietly shift to the person to discover the setting after launch.
Why this matters for expert communication
Expert content depends on a different kind of trust than generic content. The viewer evaluates the sentence or image alongside the person behind the claim, the work behind the example, and the speaker's chance to approve what left the room.
A research-guided interview gives the speaker a chance to choose the example, explain the tradeoff, add a caveat, and decide what is fit to publish.
Keep the recording so an editor can check whether a clip represents what the person said and approved.
Keep a record of sources and approvals
Design provenance into the workflow before the output looks polished. Who provided the source? Who is being represented? What did the system do? What did the human approve?
For video interviews, that means keeping the full recording as the source of record. A short clip can be useful, but it should still be connected to the fuller answer and the person who gave it. A caption can be improved, but it should not smuggle in claims the speaker did not make. A highlight can be recommended, but the final selection should remain editorial.
A practical rule for small teams
Before using AI in a public-facing asset, ask three questions. First: could this make it look like a real person said, did, endorsed, or experienced something they did not approve? Second: can we show the source trail behind the claim or clip? Third: would the person represented understand the use before it goes live?
If the answer to the first question is yes and the other two are weak, slow down. Get explicit permission. Keep the source visible. Make the AI role boring and clear.