Open Doors for Hackers: DeepSeek Left Confidential Data Exposed Online

Researchers from Wiz Research discovered an open ClickHouse database containing over a million records of confidential user information from the Chinese AI assistant DeepSeek. The publicly accessible data included unencrypted chat logs, secret keys, logs, backend, and server information.

"We conducted a reconnaissance of DeepSeek's public infrastructure and came across a database that required no authentication. This meant that anyone could access logs containing real chat messages, internal secrets, and system data," Wiz Research specialists reported.

Upon discovering the vulnerability, the Wiz team immediately contacted DeepSeek, and the company promptly restricted access, removing the database from the internet.

Privacy Policy Under Scrutiny

It was revealed that the ClickHouse database was accessible on the servers oauth2callback.deepseek.com:9000 and dev.deepseek.com:9000. The incident raises serious concerns about the data protection measures claimed by DeepSeek. According to the company’s privacy policy, all user data is stored on secure servers located in China. However, experts discovered that personal user information—including IP addresses, logs, device data, cookies, crash reports, keystroke patterns, and rhythms—remains on DeepSeek’s servers even after an account is deleted.

Wiz Research, a cybersecurity company operating since 2020, continues to monitor cloud services for vulnerabilities. Meanwhile, the DeepSeek situation once again raises concerns about the security of users' personal data in AI services.