In short
- Tea's statement: about 72,000 images from a dataset dating from before February 2024, including about 13,000 verification selfies and photo IDs.
- A second issue, reported by 404 Media on 28 July 2025, concerned more than a million direct messages, some from the week before. Tea then took its messaging system offline.
- Posts on 4chan described the storage as a publicly readable Firebase bucket. Tea's statement speaks of a legacy data storage system and does not name the product.
- The data came from old storage that was never migrated. Buckets your current app no longer uses are still reachable if their rules allow it.
- Identity documents are the worst thing to leak and the easiest to stop keeping: delete them once the verification is done.
On this page
What happened to the Tea app?
Tea is an app where women share information about men they have dated. To join, a user proves she is a woman by submitting a selfie. Tea's statement says the app also asked for a photo ID until 2023, when it removed that requirement. In late July 2025 the app reached the top of the App Store charts. 404 Media reported at the time that Tea claimed more than 1.6 million users.
On 25 July 2025, 404 Media reported that users of the forum 4chan said they had found an exposed database belonging to Tea and were posting personal data and selfies taken from it. Tea confirmed a breach the same day. NBC News reported that the company learned of the incident early that Friday, after a 4chan thread the evening before had called for a "hack and leak" campaign.
| Date | What was reported | Source |
|---|---|---|
| 25 July 2025 | 4chan users post images they say come from Tea's storage. Tea confirms unauthorised access. | 404 Media, NBC News |
| In the days after | Tea publishes its count: about 72,000 images, from a dataset dating from before February 2024. | Tea's statement, quoted by BleepingComputer |
| 28 July 2025 | A security researcher reports a second issue: private messages between users were accessible, more than a million of them. | 404 Media |
| 28 July 2025 | Tea says some direct messages were accessed as part of the initial incident and takes the affected system offline. | NBC News, BleepingComputer |
What was exposed
The figures for the first incident come from the company. Tea said about 72,000 images were accessed:
- about 13,000 selfies and photo identification images submitted during account verification;
- about 59,000 images that were publicly viewable in the app, from posts, comments and direct messages.
Tea described the cause this way: "A legacy data storage system was compromised, resulting in unauthorized access to a dataset from prior to February 2024." It added that during its early stages of development some legacy content had not been migrated to its newer system. A spokesperson told NBC News that the data had originally been stored "in compliance with law enforcement requirements related to cyberbullying prevention". BleepingComputer reported that the leaked files amounted to more than 59 GB, were shared on hacking forums, and that it had confirmed they contained driver's licences, selfies and message attachments.
The second issue concerned messages. 404 Media reported on 28 July 2025 that an independent security researcher had found it was possible to access private messages between Tea users, in a database separate from the first one, and that it had verified the findings. NBC News put the number at more than 1.1 million direct messages, dating from early 2023 to the week before the report. That is much more recent than the first dataset. According to 404 Media, the conversations covered subjects such as abortions and cheating partners, and some contained phone numbers and other details that made users identifiable.
Tea said it had engaged outside cybersecurity experts, notified law enforcement, found no evidence of access to other parts of its environment, and would offer free identity protection services to the people affected.
Was this a vibe coding failure?
The breach was widely discussed as an example of AI-generated code gone wrong. The dates do not support that for the first incident. Tea's statement places the affected data before February 2024, and the developer Simon Willison, who covered the statement, wrote that he was confident vibe coding was not to blame for that reason.
The pattern is older than any coding tool, and it is the same one behind recent cases in AI-built apps. In CVE-2025-48757, 170 of 1,645 Lovable apps tested had tables readable or writable by strangers. In the Moltbook leak, a whole Supabase database answered the public key. In each case a backend that the app talks to directly from the client was left without rules that say who may read what. The app works the same either way, so nobody notices from the inside.
The lessons for anyone who stores user uploads
- 1
Storage has its own rules
On Firebase, Cloud Storage rules are separate from Firestore rules. On Supabase, buckets have their own policies. Locking the database does nothing for the files. Firebase's guide to insecure rules names open access as the first pattern to fix: anyone who knows the project can read the data.
- 2
Signed in is not the same as allowed
The second pattern Firebase names is access for any authenticated user. If any account can read every message or every file, then anyone who signs up can. Rules have to tie each document and each file to its owner.
- 3
Old storage is still storage
Tea's dataset was legacy content that had not been migrated. A bucket, a table or a whole project that your current app no longer uses keeps answering requests for as long as it exists. Migrating to a new system does not close the old one.
- 4
Keep identity documents for as short a time as possible
A photo of a driving licence serves one purpose: a verification that takes minutes. Once it is done, delete the file and keep only the result. Data you no longer hold cannot leak. Simon Willison noted that many of the leaked photos still carried location metadata, so strip it on upload. If a law requires you to keep something, keep exactly that, in a place no client can reach.
- 5
Private messages deserve the strictest rule you have
People write things in direct messages that they tell nobody else. A messages collection should be readable only by its participants, and that rule should be the first one you test with a second account.
How to check your own storage rules
These checks run on your own project and take about ten minutes. Use a project you own, and files of your own test accounts.
Firebase
Open Storage, then Rules, in the Firebase console. Look for allow read followed by if true, or by a condition that stops at request.auth != null. Then ask the bucket as a stranger:
# Your own project only. Can anyone list the files in the bucket?
curl "https://firebasestorage.googleapis.com/v0/b/YOUR_BUCKET/o"rules_version = '2';
service firebase.storage {
match /b/{bucket}/o {
// Files look like: user/USER_ID/FILE
match /user/{userId}/{fileName} {
allow read, delete: if request.auth != null && request.auth.uid == userId;
allow create, update: if request.auth != null && request.auth.uid == userId
&& request.resource.size < 5 * 1024 * 1024
&& request.resource.contentType.matches('image/.*');
}
}
}The Firebase security rules guide has more patterns and a way to test rules with two users in the emulator. The Firebase rules checker reads pasted rules and flags the open ones. If you are counting on App Check to keep strangers out, read what App Check does and does not do first.
Supabase
In the Supabase dashboard, open Storage and look at each bucket. A bucket marked public serves any file to anyone who has its URL. For private buckets, read the policies on storage.objects: listing files and downloading them are two permissions, and a policy with a condition of true opens the bucket to everyone. The Supabase storage security guide gives per-user folder policies and explains signed URLs.
- No bucket holding private files is public or listable by strangers
- A second test account cannot open the first account's files or messages
- Identity documents are deleted after verification, or stored where no client key can read them
- Every old bucket, table and project is either deleted or locked
- Uploaded photos are stripped of location metadata before they are stored
- The checks are repeated after each deploy that touches storage
If you find your own bucket open
Close it first, by replacing the rule with one that denies everything, and accept that uploads break for an hour. Then find out what the bucket held and since when it was open. If it contained personal data, you may have duties to notify an authority and the people concerned, with deadlines that start when you become aware. The guide on what to do when your app is hacked goes through the first hour, the first day and the first week.
Tea's response shows the steps that follow at a larger scale: the affected system taken offline, outside experts brought in, law enforcement notified, identity protection offered to those affected. A solo maker does the same things in smaller form.
What Keybrum tests on storage, and what it does not
Keybrum is the automated pentest we build for apps made with AI. It starts from the address of a web app, reads the JavaScript it serves, and attacks the backend it finds as a stranger and as newly created accounts.
- Firebase Storage. It lists the bucket and the folders named in the app's code, flags files that look like receipts and identity documents, and downloads one byte of one file as proof. File contents are not stored.
- Supabase Storage. It lists the buckets named in the code, walks their folders, flags identity and financial documents, and downloads one byte. It also tries file paths found in readable database rows, for buckets that cannot be listed.
- Collections and tables. It reads Firestore collections and Supabase tables with no account, then as a new account. The second pass is the one that catches rules open to any signed-in user.
The limits matter here. Keybrum works from what the current web app names, so a legacy bucket that no code refers to any more, like the one in Tea's account of the incident, would not be found: your own inventory of old storage is the only check for that. It does not unpack an iOS or Android app. It reports what the backend answers, and does not review how long you keep documents. It only attacks apps whose owner proved ownership.
A one-time report costs €49. The Maker plan, €29 a month excluding VAT for 5 apps, attacks again on every deploy.
Questions
What was leaked in the Tea app data breach?
According to Tea's statement, about 72,000 images: about 13,000 selfies and photo IDs from account verification and about 59,000 images from posts, comments and direct messages. A second issue reported on 28 July 2025 concerned more than a million private messages between users.
When did the Tea app breach happen?
Tea confirmed unauthorised access on 25 July 2025, after images were posted on 4chan. The second issue, concerning direct messages, was reported by 404 Media on 28 July 2025. Tea said the first dataset dated from before February 2024.
Was the Tea app using Firebase?
The 4chan post that publicised the leak described an unsecured Firebase storage bucket, and 404 Media and BleepingComputer reported that claim, attributed to the poster. Tea's own statement refers to a legacy data storage system without naming the product.
How many messages were exposed?
404 Media wrote that it had obtained more than one million messages. NBC News reported more than 1.1 million direct messages, dating from early 2023 to the week before the report. Tea said some direct messages were accessed as part of the initial incident.
Was the Tea app breach caused by vibe coding?
The first dataset predates February 2024 according to Tea, and the developer Simon Willison argued on that basis that vibe coding was not to blame. The underlying problem, storage readable without the right rules, is older than AI coding tools and also appears in apps built with them.
How do I check if my Firebase Storage is public?
Read your Storage rules in the Firebase console and look for allow read: if true or a rule that only checks request.auth != null. Then request the bucket's file list from a terminal without signing in. A permission error means the rule held.
Sources
- 1.Women Dating Safety App 'Tea' Breached, Users' IDs Posted to 4chan, 404 Media
- 2.A Second Tea Breach Reveals Users' DMs About Abortions and Cheating, 404 Media
- 3.Hackers leak 13,000 user photos and IDs from the Tea app, designed as a women's safe space, NBC News
- 4.Tea app leak worsens with second database exposing user chats, BleepingComputer
- 5.Official statement from Tea on their data leak, Simon Willison
- 6.Fix insecure rules, Firebase Docs