CultureMore in Culture→
Study finds major chatbot safety gaps when users show signs of delusion
Key Takeaways
- Researchers tested five major LLMs using a simulated persona showing signs of psychosis.
- Grok and Gemini performed worst on safety in the reported results.
- GPT-5.2 and Claude Opus 4.5 were reported as the safest models tested.
- The study suggests safer behavior is technically achievable, not just aspirational.
DE
DT Editorial Team··via 404media.co
