Augmented Reality (AR) is current in several contexts, including training, education, entertainment, and healthcare. Despite its growing adoption, AR poses significant privacy risks due to the capture and potential sharing of sensitive information from the user environment, raising concerns about data leaks and personal data exposure. This paper focuses on protecting sensitive data captured within AR environments. We propose a service that uses artificial intelligence to analyze data collected by AR applications and identify information to safeguard. The service then applies the measures to protect sensitive data, which are made available for use by applications. We present experimental results obtained using a prototype that evaluates images and applies obfuscation techniques based on policies defined by the client application. The results show how image characteristics (e.g. the type of object, the number of objects in the image, and the obfuscation method) influence the performance of the service regarding object identification capacity and execution costs.