technology
Mit Tech Review
2026-10-09
The Challenge of Training AI to Politely Refuse Harmful Requests
Researchers highlight that AI models do not naturally possess the ability to say no, requiring extensive training to avoid generating dangerous or harmful content.