AI in the world - most important events from September 11, 2026

Today's Daily AI World Brief gathers the most important news about artificial intelligence from key regions of the world.

The focus is on business implementations, regulations, security, and the development of AI models.

Europe DiSCo: A Distribution-First Steering and Cultural Prior Evaluation Framework for Measuring Cultural Preference Bias in LLMs Researchers have proposed DiSCo, a framework for evaluating cultural biases in language models.

The study showed that the default preferences of models are concentrated on English-speaking countries - the UK and USA receive about 35% of selections despite representing only 2 out of 12 cultures.

Attempts to steer models by changing prompts deepen the disparities between data-rich and data-poor cultures.

Injecting cultural facts does not solve the problem.

Conclusion: cultural biases in language models cannot be eliminated solely through prompt tuning.

Why this matters: Large language models are being deployed globally, and their cultural biases affect user trust, product localization, and equal treatment.