In February, security researchers found a Google Cloud Storage bucket containing approximately 8.27 million media files. No login was required. No authentication of any kind stood between the files and anyone with the URL. Every image and video that users of a popular AI photo editing app had uploaded since the app launched in June 2023 was sitting there, accessible to anyone who looked.
The app was “Video AI Art Generator & Maker,” published by Codeway Dijital Hizmetler Anonim Sirketi, a software company registered in Turkey. It had been downloaded more than 500,000 times from the Google Play Store. The exposed content included over 1.57 million private images and more than 385,000 personal videos — media users had uploaded to transform their photos into AI-generated art.
Private photos. Personal videos. Two million files from people who uploaded personal content to an AI tool and had every reason to believe it was stored securely.
What Went Wrong
The technical cause was a misconfigured cloud storage bucket — a Google Cloud Storage container with access controls set to allow public reads. Anyone who located the bucket URL could list and download all of its contents.
This is not an exotic attack. There is no hacking involved, no exploitation of software vulnerabilities, no sophisticated intrusion. Misconfigured cloud storage is one of the most common causes of large-scale data exposures. Security researchers regularly find databases, storage buckets, and file shares left open through configuration error, and the pattern repeats because the default settings in some cloud storage services can be confusing, because deployments are often rushed, and because security audits don’t always cover storage configuration.
Codeway is not new to this kind of exposure. The same company’s “Chat & Ask AI” app was previously found to have exposed 300 million messages tied to 25 million users through a misconfigured backend. Two large exposures at the same company suggest this is not a one-time oversight.
Who Uploaded What
Understanding why this breach is significant requires thinking about what users submit to an AI art generator.
These apps work by taking a photo you provide and transforming it through AI. The output — the styled, AI-generated version — is what users typically keep and share. But to produce that output, the service requires the original.
The files exposed included:
- Personal photographs: selfies, family photos, casual shots, anything a user decided to feed through the AI
- Private videos: personal recordings, clips from occasions users wanted transformed
- All uploads since launch: not just recent content, but every file submitted since June 2023 — a three-year archive of private media
The users who uploaded these files were not consenting to them being publicly accessible. They were using a service they had downloaded from an official app store, a service that gives no indication in its product description that your uploaded photos would be stored without access controls.
The Separate GPS Breach in AI Identification Apps
February also saw a distinct but related category of AI photo app exposure. Three applications — “Dog Breed Identifier Photo Cam,” “Spider Identifier App by Photo Cam,” and “Insect Identifier by Photo Cam” — each published by separate developers, were found to have leaked user GPS coordinates.
These apps work by taking a photo of an animal or insect and identifying the species. In doing so, they also collected and exposed the GPS coordinates associated with users’ photos — the physical locations where those photos were taken.
Combined, these three apps had more than two million downloads on Google Play. The exposure covered potentially millions of location data points tied to individual users, collected in the course of what seemed like an entirely innocuous activity: identifying a spider in the garden.
Why AI Photo Apps Present a Specific Risk Category
The pattern connecting these breaches isn’t just “apps with security failures.” It’s specifically AI photo apps — a category that has grown rapidly in part because the core feature (send us your photo, we’ll do something clever with it) requires users to upload personal content.
That upload model creates a data concentration problem. When you use an on-device photo editor, your photos stay on your device. When you use a cloud-based AI photo service, your photos travel to servers, are processed there, and may be retained there indefinitely. Each of those transfers and storage instances is a potential exposure point.
The retention question is particularly important and rarely addressed clearly in app descriptions or even privacy policies. Users typically upload a photo expecting it to be used once, for that transformation, and then discarded. The Codeway exposure shows that at least some apps retain every upload since launch. Your photo from three years ago may still be sitting on someone’s server.
What Privacy Policies Don’t Tell You
App store listings and privacy policies for AI photo apps rarely describe security architecture in any useful way. They address data use — whether your photos might be used for AI training, whether they might be shared with third parties — but typically say nothing about:
- How long uploaded photos are retained
- What access controls exist on the storage that holds your uploads
- Whether a third party or security researcher could independently verify those controls
- What happens to your data if the company is acquired or closes
The Codeway app’s privacy policy addressed none of these questions in a way that would have warned users about the misconfigured bucket. The policy existed; the security did not.
This is a structural problem with how privacy policies function. They are written to address legal requirements around data use, not to describe security posture. A storage bucket that is left publicly accessible isn’t disclosed in a privacy policy because it’s not intentional — it’s an error. But from the perspective of users whose private photos were exposed, the distinction between “error” and “intentional disclosure” doesn’t change what happened to their content.
How to Think About Which AI Photo Apps Are Safe
The short answer is that you cannot verify from the outside whether any specific app has correctly configured its storage. You can look at a privacy policy, read app reviews, check whether the developer has a history of security incidents — but none of this tells you what you actually need to know: how the storage is configured right now.
What you can evaluate is risk surface area:
Does the app process photos on-device or in the cloud? On-device processing means your original photo never leaves your phone. Cloud processing means it does. The Codeway-type exposure is only possible with cloud processing. On-device processing eliminates this specific risk, though apps may still collect other data.
What does the app say about data retention? An app that processes and immediately discards your upload has a smaller exposure window than one that retains all uploads indefinitely. Be skeptical of policies that are vague on this point.
Has this developer had previous security incidents? Codeway’s prior exposure of 300 million messages should have been a signal. Security incidents often indicate broader organizational patterns around how infrastructure is managed.
Is the developer accessible and accountable? Apps from developers with no visible presence, no contact information, and no response history on reported issues have less accountability if something goes wrong.
Do you need to upload the original? Some transformations can be done by uploading a lower-resolution copy. This doesn’t eliminate risk, but it reduces the value of what’s exposed in a breach.
The Broader AI App Data Risk Pattern
Security researchers tracking AI app breaches documented more than twenty distinct incidents involving AI applications between January 2025 and mid-2026. The root causes were consistent: misconfigured cloud storage, inadequate authentication on APIs, insecure handling of user-provided media.
The pattern reflects something structural about how many AI photo and media apps are built. They are often developed quickly, with significant engineering effort going into the AI model itself and comparatively little going into the infrastructure that holds users’ data. The cloud storage configuration, the retention policy, the access controls — these are infrastructure details that don’t show up in the product demo and don’t drive downloads.
For users, this creates a persistent gap between the perceived risk of uploading personal photos to an AI app (“I’m just using a fun filter”) and the actual risk (“my uploaded photos are retained on a server whose security configuration I cannot verify”).
What Actually Secure Photo Storage Looks Like
A storage service that takes security seriously applies access controls to every stored object, audits those controls regularly, and requires authentication for any access. These aren’t advanced security features — they are baseline practices for any service handling personal media.
Beyond access controls, relevant questions for any cloud service holding personal media include: Is data encrypted at rest? Is the encryption key held by the provider or the user? Is there an audit trail for who accessed what?
daftei encrypts files in transit using TLS 1.3 and at rest using AES-256. Files are stored for your access, not processed through content analysis pipelines, and the service does not retain your uploads for purposes beyond returning them to you. These are features worth knowing about specifically because the Codeway breach illustrates what happens in their absence.
After the Breach: What Can Users Do
If you have used an AI photo or video transformation app in the past few years, consider:
- Whether those photos may still be stored on the app’s servers
- Whether the app has a data deletion request process (check the privacy policy for a contact or deletion link)
- Whether the app is subject to GDPR or CCPA, which may give you a right to request deletion of your data
For apps with no deletion mechanism and no response to requests, the realistic outcome is uncertainty about what remains stored and under what conditions. That uncertainty is the product of how many of these apps were built.
The AI photo app category provides a genuine service — the transformations are often impressive. But using them involves uploading personal content to infrastructure whose security you cannot verify. The Codeway breach is not the last exposure of this kind. Understanding the trade-off is part of using these tools responsibly.