DeepSeek

DeepSeek, a Chinese competitor to OpenAI’s ChatGPT, received massive public attention and soared to the top of the App Store download charts when in launched recently. Here are some of the security-related events that subsequently occured.

• Harmonic Security took a look at the data privacy concerns around the Chinese AI company, highlighting vague statements about data retention within the People’s Republic of China. The AI security firm concluded that very few (0.21%) of its customer’s users were actually using DeepSeek though. harmonic.security  
• DeepSeek limited signups amid a sudden wave of interest and in response to what it described as “large-scale malicious attacks on DeepSeek’s services”. theregister.com 
• Lots of examples have been shared on social media of DeepSeek refusing to answer questions about topics the Chinese Communist Party deems sensitive, such as the Tiananmen Square Massacre. An analysis by PromptFoo of 1,156 prompts found that these “canned refusals” were given 85% of the time and were reasonably easy to circumvent, suggesting, they say, that the censorship is more of a “crude, blunt-force” implementation rather than deeply baked into the reasoning model itself. arstechnica.com 
• The Chinese company appears to have pretty sloppy security engineering practices: Wiz security researchers found a publicly accessible database containing “a significant volume of chat history, backend data and sensitive information, including log streams, API Secrets, and operational details,” within ‘minutes’ of scanning DeepSeek’s infrastructure. The HTTP interface to the database allowed Wiz to run a SHOW TABLES; query, returning all the accessible tables. The log stream data may have included plaintext passwords and chat history. DeepSeek promptly fixed the issue after being notified. theregister.com 
• Italy blocked DeepSeek over privacy concerns after the company told the Italian data protection regulator that it did not fall under the purview of GDPR. therecord.media