Wiz researchers find sensitive DeepSeek data exposed to internet

A publicly accessible ClickHouse database tied to AI company DeepSeek exposed over a million lines of sensitive internal data — including plaintext user chat logs, API keys and cryptographic secrets, server directory and operational metadata, and internal API references — discovered by Wiz researchers via open ports (8123 and 9000). DeepSeek secured the database after notification; researchers warn attackers could have extracted files or leveraged secrets for privilege escalation or corporate espionage, and separate Kela analysis highlights the R1 model's susceptibility to jailbreaks, underscoring broader security risks in rapidly deployed AI services.