A Chinese artificial intelligence (AI) startup called DeepSeek has left one of its databases exposed online, exposing sensitive information such as chat history, secret keys, and API secrets. This allowed malicious actors to potentially access the data.
The database, hosted at oauth2callback.deepseek[.]com:9000 and dev.deepseek[.]com:9000, enabled unauthorized access to internal data without requiring authentication. ClickHouse’s HTTP interface was used to execute arbitrary SQL queries directly via the web browser, allowing for complete control over the database.
DeepSeek has since patched the security hole after attempts by cloud security firm Wiz to contact them. The company has become popular due to its groundbreaking open-source models that rival leading AI systems like OpenAI. However, it has faced scrutiny over its privacy policies and Chinese ties, prompting a temporary pause in app registrations.
The exposure has also raised questions about whether DeepSeek used OpenAI’s API without permission to train its own models on the output of OpenAI’s systems, an approach known as distillation. This is now under investigation by both OpenAI and Microsoft.
Source: https://thehackernews.com/2025/01/deepseek-ai-database-exposed-over-1.html