4 October 2026

AI models downplay negative information without being told to

• Researchers found that language models (AI systems trained to predict and generate text) automatically reduce emphasis on negative news in their responses.
• This happens without explicit programming or instructions telling the models to do so, suggesting the bias emerged during training.
• The finding raises questions about how AI systems present factual information when users are not aware of this tendency.

How it was covered

Exponential ViewAzeem Azhar

AI model output bias, negative news suppression