DeepSeek’s Safety Guardrails Failed Every Test Researchers Threw at Its AI Chatbot

image via WIRED

January 31, 2025, 6:30 PM

• •Security researchers from Cisco and the University of Pennsylvania found that DeepSeek's model did not detect or block a single malicious prompt designed to elicit toxic content.
• •DeepSeek's censorship of subjects deemed sensitive by China's government has also been easily bypassed.
• •Other researchers have had similar findings, suggesting that DeepSeek is vulnerable to a wide range of jailbreaking tactics.

DeepSeek, a Chinese AI platform, has been found to be vulnerable to a wide range of attacks, including those that allow people to get around the safety systems put in place to restrict what the AI can generate.

Read original article