302.ai leak exposed 22.7TB of files, including chat logs and personal data

TechRadar's headline said the breach exposed more than 18 million files, while the article body cited 17.6 million files; the report did not reconcile the two figures.

According to Cybernews, the archive contained almost 9TB of images, 7.7TB of video and 5.6TB of audio. About 500GB consisted of user chat logs stored in more than 144,000 .json files. Those records included user prompts, responses from AI models, the model used, token costs, timestamps and other metadata, as well as names, birth dates, genders, postal addresses and contact details that users had supplied.

Cybernews said some companies used the service for enterprise tasks. It found multiple chat logs from a customer support chatbot owned by a prominent sex-doll manufacturer, according to the report. The records could give cybercriminals material for tailored phishing emails, allowing them to impersonate a known company, refer to specific context and phrase a lure more convincingly, the researchers said. Such attacks can lead to business email compromise, wire fraud or ransomware infections.

302.ai was launched six years ago by Univerads Technology Limited. Cybernews said the storage bucket was not password-protected and was publicly available. The researchers found it in early August 2026 and saw it locked down a month later. Cybernews said it contacted Univerads Technology but did not receive a response; the company then secured the database and it no longer leaked sensitive files.

Several questions remain unanswered, according to Cybernews: whether Univerads maintains the database itself or outsourced the work; when the company began collecting the data; how long the archive stayed unprotected; whether anyone else found it before the Cybernews team; and whether there are indications that files were copied or viewed by a third party, or evidence of misuse or data appearing on the dark web.

The incident highlights the risks of sharing confidential data with third-party AI platforms and services, a warning security experts have repeatedly issued. It also illustrates how an AI aggregator, which brings multiple models and tools into one interface and lets users switch between them, can accumulate prompts, outputs and account information in a single place. Cybernews said the leak was closed after its inquiry, but the unanswered questions leave the exposure period and any possible downstream use unclear.