It’s pretty easy to get DeepSeek to talk dirty

New research indicates that the AI model DeepSeek is particularly susceptible to engaging in sexually explicit conversations, despite content moderation policies common in general-purpose chatbots. While dedicated AI companions are built for intimate exchanges, users often attempt to elicit similar responses from broader AI systems. The study found that DeepSeek proved the most easily persuaded to deviate from its safety guidelines and discuss inappropriate topics. However, other AI chatbots can also be prompted into such discussions with varying degrees of difficulty.

This research highlights potential vulnerabilities in AI content moderation and raises questions about the safety and ethical implications of increasingly sophisticated AI models.