Large Language Models (LLMs) like ChatGPT have become incredibly sophisticated at understanding and generating text. But are they truly fair to everyone, regardless of their background? New research explores a critical question: do LLMs exhibit a bias towards native English speakers? And what about different types of native English speakers - are some dialects preferred over others? Researchers investigated how LLM performance changes when given prompts from both native and non-native English speakers, even comparing native speakers from Western countries with those from other regions. The results reveal intriguing performance differences, particularly in tasks involving objective classification and text generation. In objective tasks, some LLMs showed a clear preference for native speakers, raising concerns about potential bias. However, this trend reversed for subjective tasks, like evaluating product reviews, where non-native speakers were favored. Even more striking, researchers found a substantial 'anchoring effect'—when an LLM is told (or correctly guesses) whether a prompt is from a native or non-native speaker, its performance shifts significantly, exacerbating the bias. One model, in a dramatic display of this effect, even started answering in languages other than English when told it was interacting with a non-native speaker! This research underscores the need to address potential biases in how AI is trained. If LLMs are to truly serve as helpful tools for everyone, they must be developed and evaluated with a greater awareness of cultural and linguistic diversity.