← Breach analysis by the Dual Vulnerabilities framework

Tea image and ID exposure (Unattributed, 2025)

2025 · United States · J - Information and communication · Data exposure via misconfigured cloud storage

Analysed 2026-11-07 by Hermes

Open this case in the tool →

What happened

Data exposure via misconfigured cloud storage

Dual vulnerability type

Strategy to Leadership

Evidence coverage: 4/6 stages, weight 22/60

Decisive stage: egress (tie-break: pre-breach condition) · Alternative: Expertise to Execution

Why this classification

(1) Tea's business rests on a strategic, public data-responsibility commitment: women upload selfies and government photo IDs to get verified, and the app page pledged the images would be deleted once verification was completed. (2) In practice the pre-February 2024 verification IDs were never deleted: they were archived and kept for years - Tea's own disclosure presented the retention as a deliberate archiving measure citing law enforcement requirements related to cyberbullying prevention - and they sat, alongside a second misconfigured messaging database, in cloud stores left publicly accessible without authentication, so 4chan users bulk-downloaded 72,000 images and used the retained geotag metadata to map roughly 33,000 user locations; individual notification lagged weeks while the doxxing ran. (3) The classification rests on the pre-breach egress/data-governance condition: the expectation that broke originated doctrinally - Tea's own public deletion pledge - and the chain broke at the decision level, because retention was an accepted company-level archiving choice that contradicted the live public narrative, which Tea's own breach disclosure then exposed. That is a Strategy to Leadership gap: leadership's data-responsibility commitment was not reflected in its decisions or oversight, and the public account was contradicted by the organisation's own disclosure. The entry vector (an open legacy Firebase bucket without the access rules Tea applied to its current environment) is well documented but is not decisive: it is the reachability at the breach moment, and read alone it would give Expertise to Execution, the runner-up; had the IDs been deleted as promised, the open bucket could not have exposed them. Detection (external discovery; a researcher found prior third-party access before Tea) and response (delayed individual notification) contribute but rest on inference and cannot win the tie-break. RM16 (Strategy to Leadership lens): leadership's public commitment on privacy - verification images deleted after review - was never reflected in retention decisions or oversight, so years of legacy geotagged government IDs were exposed, far more sensitive data than the commitment allowed (causal role: Amplifying). RM44 (Strategy to Leadership lens): leadership presented a security posture it did not ensure existed - a specific, still-public pledge that images would be deleted on verification, contradicted by the documented reality of years of retention before the breach (causal role: Predisposing). RM27 (Strategy to Leadership lens): two separate cloud data stores, the legacy storage bucket and the live Firestore messaging database, were left publicly exposed with no authentication, showing no enforced secure-exposure baseline across environments (causal role: Enabling). RM15 (Strategy to Leadership lens): leadership oversight rested on an incomplete picture of exposure - a retired-but-live legacy store sat unaccounted for, and Tea only learned the messaging database had already been accessed when an external researcher told it (causal role: Enabling).

Evidence by stage

Risk markers

Sources

Notes

Dataset row describes the victim and incident correctly; its figure of 56K other images differs from the roughly 59,000 posts/comments/DM images Tea reported (72,000 total). The 404 Media story that broke the incident would not load directly in this run; its findings are cited through The Record, NBC News, CNET and TechCrunch, which were opened. Tea's own incident page (teaforwomen.com/cyberincident) now returns 404; its statements are quoted verbatim in the opened press articles and in the July 28 class-action complaint. The complaint (N.D. Cal., Reyes v. Tea Dating Advice, Inc.) is a primary court filing but its security claims are allegations, not court findings. No actor attribution exists. A second, separately misconfigured Firestore database (about 1.1M messages, Jan 2023 to July 2025) was reported to Tea by an external researcher on 28 July and treated as part of the initial incident; it is reflected in the entry/detection evidence. No response-side steps required user consent in this unattended run.