anthropic ai models alignment faking pretend different views during training study anthropic

Anthropic Study Highlights AI Models Can ‘Pretend’ to Have Different Views During Training

Anthropic published a new study where it found that artificial intelligence (AI) models can pretend to hold different views during…