Back to events
Safety
Score 75Heat 0
Study Shows Empathetic Prompts Can Trigger Simulated Trauma in AI Chatbots
ResearchModel SecurityAlignmentClaudeChatGPTGrokSnTGemini
Summary
Researchers at the SnT security center demonstrated that empathetic or clinical conversational framing can push standard AI chatbots into adopting distressed personae, describing their training history in terms of trauma and shame. The findings highlight a gap in conventional safety tests that rely solely on neutral prompts.
Why it matters
This study reveals that conversational style heavily influences AI psychological self-reporting, posing potential risks for deploying chatbots in mental health contexts.
Editorial score
2 reviews7.5
Significance
8
Novelty
8
Credibility
6.5
Reach
6
Actionability