security

AI Photo Apps Are Leaking Your Images. Here's What Happened.

A February 2026 investigation found AI photo editing apps exposing 8.27 million media files. The cause was cloud misconfigurations — and it's a widespread pattern.

In February 2026, security researchers published findings on a category of apps that had become enormously popular: AI-powered photo editors, makeover tools, and beauty filter apps on the Google Play Store. The investigation found that several of these apps had misconfigured cloud storage backends that left user photos publicly accessible — no authentication required.

One app alone had exposed approximately 8.27 million media files totaling more than 12 terabytes. PetaPixel reported the photo count at 1.5 million images. A separate app, IDMerit, had exposed Know Your Customer (KYC) verification records belonging to users in at least 25 countries — including government ID photos submitted for identity verification.

This wasn’t a sophisticated attack. No one brute-forced a password or exploited a zero-day vulnerability. The files were simply sitting in cloud storage buckets that had been configured to allow public access, where anyone who knew the storage address could retrieve them.


Why AI Photo Apps Are a High-Risk Category

AI photo editing apps sit at an intersection of several properties that make them particularly risky from a data-security perspective. Understanding why requires looking at how these apps work, not just what they claim.

They need your photos to function. Unlike apps that process data locally on your device, AI photo editing apps typically upload your images to cloud servers for processing. The AI model runs server-side, the result gets sent back to you, and the original file may or may not be deleted from the server afterward. “May or may not” is doing a lot of work in that sentence.

They’re often built by small teams with limited security expertise. The AI photo editing category saw explosive growth from roughly 2023 onward, with hundreds of apps entering the space. Many were built quickly by small development teams focused on the AI feature, not the infrastructure that stored user data. Cloud storage misconfiguration is a common result of teams that move fast without security review.

They request broad permissions. To access your camera roll, these apps request full photo library access — not just the photo you’re currently editing. For developers who build retention into their backend, this means the server can receive copies of photos that users didn’t specifically intend to upload.

User trust is high. AI photo apps are positioned as helpful tools. They promise to make your selfies better, remove wrinkles, swap hairstyles, or restore old photos. The emotional stakes — improving something personal about yourself — create willingness to grant access without scrutinizing what happens to the result.


The KYC Dimension

The IDMerit exposure introduced a separate category of risk: government-issued identity documents.

KYC — Know Your Customer — is a verification process that apps in finance, age-restricted content, and cryptocurrency have adopted at scale. The user takes a photo of their passport or driver’s license, submits it to the app, and a verification service confirms their identity. For many users this feels routine: the government requires age verification, the app requires KYC, you comply and move on.

What the February 2026 exposure revealed is that those ID photos, along with the personal data extracted from them — full legal name, date of birth, ID number, home address — were stored in an external vendor’s database that had inadequate access controls.

An exposed government ID photo is a different category of risk than an exposed selfie. It contains everything needed for identity theft, fraud, and account takeover. It’s not possible to issue yourself a new passport because someone has a photo of your current one. The exposure is permanent in a way that credit card fraud is not.

The reach of that exposure was global: KYC records belonging to users in 25 countries, many of whom had no idea their identity documents were stored by an app they’d briefly used months earlier.


How Cloud Misconfiguration Happens

The technical cause of these exposures is consistent across incidents: cloud storage buckets or blob containers configured to allow public access, when they should allow only the application backend to read and write files.

Major cloud providers — Amazon Web Services, Google Cloud, Microsoft Azure — all provide simple controls to prevent public access. AWS S3, for example, has a “Block Public Access” setting at the account level that prevents any bucket from being made public, even accidentally. Google Cloud has equivalent organization-level policies.

These controls exist because misconfiguration is common. Developers creating a bucket might set it to public access for testing, intend to restrict it before launch, and forget. A default configuration in an older SDK version might allow public access. A teammate might change a setting that another teammate doesn’t know about.

When these mistakes happen at large companies with security teams, they’re often caught in code review or automatically flagged by security tooling. When they happen at small app development teams, there’s no one checking.

A 2026 survey by Mysterium VPN found approximately 19.6 billion files sitting in publicly accessible cloud storage buckets across major providers. The AI photo editing category contributes to that number, but it’s not unique — the pattern of undiscovered misconfiguration appears across healthcare, education, finance, and consumer software.


What App Stores Can and Can’t Do

Google’s response to the February 2026 report noted that it had rejected nearly two million apps and blocked more than 80,000 developer accounts in 2025. Apple tightened its App Review Guidelines in late 2025 to require explicit user permission before apps can share personal data — including photos — with third-party AI services.

These are meaningful steps. But app store review doesn’t include dynamic security testing of an app’s cloud backend. A reviewer can check whether an app asks for appropriate permissions, discloses its data practices, and functions as described. They can’t check whether the storage bucket the app uploads your photo to is publicly accessible.

The gap between app store policy and app security reality means users can’t rely on the presence of an app on a major platform as a signal about the security of its backend infrastructure.


Reducing Your Exposure to AI Photo App Risk

Given that the category has a demonstrated history of data exposure, a few practices reduce your risk without requiring you to stop using AI photo tools entirely.

Check the developer’s privacy policy before granting camera access. Specifically look for: how long the app retains uploaded images, whether images are processed locally or server-side, and whether any data is shared with third parties for processing. Many apps in this category share with external AI model providers whose retention policies may be even less clear than the app’s.

Use limited photo access where available. On iOS, the “Selected Photos” permission option lets you grant access to specific images rather than your full library. On Android, similar granular options are available in Android 14 and later. Using limited access prevents an app from accessing photos you didn’t specifically intend to share.

Avoid uploading images that contain sensitive context. Profile photos, selfies taken at home, and photos that appear in a larger set with location data are all richer targets than a standalone cropped image. An AI editing app doesn’t need your entire camera roll or images that reveal your home address through background details.

Be more cautious about apps requiring ID verification. Any app that asks you to upload a government-issued ID photo is creating a record that, if exposed, has identity theft implications. The risk calculus for using an ID-requiring app is different from the risk of using a photo filter app.

Delete your account when you stop using an app. Photo editing apps don’t always delete your uploaded images when you stop using the app. Actively deleting your account — and verifying that your data has been deleted, not just deactivated — removes files that would otherwise sit in a backend indefinitely.


The Broader Pattern

The February 2026 findings fit a pattern that security researchers have documented across mobile app categories for years. The issue is not unique to AI photo apps, and it’s not unique to small developers. Misconfigured cloud storage has exposed medical records, legal documents, financial data, and personal photos at organizations of every size.

What makes AI photo apps a particularly concentrated risk is the combination of: broad photo access permissions, server-side processing that requires uploading files, rapid category growth without commensurate security maturity, and the nature of the data involved — personal photos that users often consider private.

For personal photos and documents that have genuine private value, the relevant question isn’t whether any given app will be breached. It’s whether the app’s security practices limit the consequences if a misconfiguration occurs. An app that processes locally and doesn’t retain copies has a much smaller exposure surface than one that uploads your full library to a server and retains images indefinitely.

Where you choose to store things you consider private matters. A misconfigured bucket can expose what’s in it. Keeping your personal files in a service that encrypts at rest, limits third-party sharing, and doesn’t retain your data as a byproduct of processing is not a theoretical benefit — it’s the difference between what was in the February 2026 exposures and what wasn’t.

Your memories deserve better than an ad platform.

Try daftei free →
← All posts