Researchers have discovered that when calling an AI model's API, the model being used may not always match the one specified. Instead, the model may be switched to a different version, potentially altering the output. This issue was observed when a request was classified as sensitive and was then handled by a different set of weights. The incident highlights the need for greater transparency in AI model usage and potential consequences for users who may not be aware of the model change. This matters because it raises questions about the reliability and consistency of AI-generated content.