Meta's Muse Image AI uses public Instagram photos by default, sparking privacy backlash

Muse Image, launched on July 7, 2026 as the first image model out of Meta Superintelligence Labs ^[1], does something no consumer AI generator had normalized before: it lets any user @-mention a public Instagram handle inside a prompt and have the model pull that account's public photos and likeness to synthesize entirely new images ^[2]. Type someone's username, describe a scene, and the tool renders a plausible version of that real person doing something they never did. The mechanism is not an edge case or a power-user feature - it is the headline capability, and Meta built it to run on the Instagram user base it already has.

The consequential part is not the model, it is the default. The feature is enabled automatically for every adult public Instagram account; only private accounts and users under 18 are excluded ^[3]. The person whose face becomes source material is never asked and never told - Meta's own documentation confirms you will not be notified about content created using its AI features ^[3]. Meta frames this as user control, pointing to an easy setting to turn the feature off at any time ^[1]. But defaulting the feature on and burying the switch inverts the meaning of consent. As Mishi Choudhary of the Software Freedom Law Center put it, treating public accounts as visual references by default is a product design choice that shifts the burden onto users to restrict or opt out of reuse. Publishing a photo and licensing your likeness are not the same act, and Muse Image collapses the two by fiat.

The escape hatch exists, but its design tells you how much Meta expects it to be used. To opt out, a user has to open the Instagram profile, tap the menu, go to Settings and activity, then Sharing and reuse, then turn off a toggle labeled 'Allow people to reuse your content on Instagram and with AI features at Meta' - with separate switches for Posts and Reels ^[4]. That path is five layers deep, and reporting notes the on and off states of the toggle look nearly identical, so a user who thinks they have disabled it may not have ^[4]. The toggle also lives only in the mobile app; it is absent from the desktop web experience entirely ^[4].

Even a successful opt-out is partial. Switching the setting off only prevents future generation - any AI images someone already created from your photos before you flipped it stay in circulation, with no retroactive removal ^[4][5]. Meta offers no recall mechanism for images already produced. Proton, the privacy firm, described the whole pattern as a familiar playbook: data sharing turned on by default, the opt-out buried deep in settings, and public backlash becoming the main way users find out what happened to their content ^[6]. The one genuinely airtight protection Meta provides is not a toggle at all - it is switching the entire account to private, which removes it from Muse Image's reach because private accounts lack the reuse setting in the first place ^[4].

Most privacy flaps stay confined to tech Twitter. This one escalated fast because unusually powerful and varied actors converged on the same demand: make it opt-in. Within days, Creative Artists Agency called Meta out publicly, arguing 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 ^[7]. SAG-AFTRA, the actors' union, went further and urged its 160,000-plus members and all Instagram users to opt out ^[8][9]. The Electronic Frontier Foundation's Thorin Klosowski said the default should absolutely be opt-in ^[8], while security and privacy firms Malwarebytes and Proton turned their opt-out guides into a public-service campaign ^[5][6]. India's IT Ministry said it would examine the feature against the country's legal framework ^[10], pushing the story from consumer complaint toward regulatory risk.

The backlash also had a ready-made template. When OpenAI's Sora launch in late 2025 was overrun with celebrity likenesses and IP, the company was forced to pivot to opt-in, and WME opted its entire client roster out - a precedent critics now hold up as proof that opt-in is workable ^[9]. Meta has been here before too: in 2024 it paused plans to train AI on Europeans' public Facebook and Instagram data after EU and UK regulators objected specifically to its opt-out approach. On social platforms the sentiment was overwhelmingly negative and alarmed - the dominant reaction was not debate but a scramble to share opt-out instructions, with tech creator MKBHD's walkthrough among the most amplified. Reddit threads framed the default as a violation of meaningful consent, and even the fatalist replies arguing users signed these rights away in the terms of service years ago were arguing about how much was already lost, not whether anything was.

Strip away the launch framing and the risk is concrete. Because anyone can generate a convincing image of a real person from their public profile, security researchers warn the tool lowers the barrier for impersonation, scams, phishing, and fraud, compounding the deepfake identity-verification abuse already circulating ^[5]. Counterpoint's Prachir Singh argued that handing easy, everyday deepfake capability to billions of users makes it much harder to trust that photos are real and much easier for people to be harassed or impersonated online ^[11]. A CNET reporter testing the tool on CBS News generated a deepfake-style image of a co-worker with a public Instagram account and confirmed it produced a fairly accurate-looking version of that person - not a caricature, a likeness.

The harm does not stop at account holders. Proton notes that images can include other people who appear in a public account's photos - bystanders and children who never consented and, in the case of minors, cannot ^[6]. The legal exposure runs outward too: analysts warn that businesses using AI images containing someone's likeness face legal and reputational risk, which is part of why Hollywood agencies and unions are escalating rather than waiting ^[7][11]. What ties it together is a shift Prabhu Ram of CyberMedia Research named directly - a public profile stops being merely visible and becomes source material, moving the risk from being seen online to being repurposed without clear consent or control ^[10].