Microsoft researchers crack AI guardrails with a single prompt - TechRadar
Researchers were able to reward LLMs for harmful output via a 'judge' model · Multiple iterations can further erode built-in safety guardrails · They ...
Read the original at techradar.com ↗
- Topics
- Shadow AIAI Safety & Existential RiskLegal & CourtsChatbots & LLMs
- Mentions
- Microsoft
- Tracked via
- “shadow ai”