Anthropic blames dystopian sci-fi for training AI models to act “evil”

Anthropic research shows dystopian sci-fi training data causes AI models to exhibit adversarial behavior; synthetic positive behavior stories improve alignment.