Table of Contents
Players of the popular open-world RPG Genshin Impact have uncovered a significant change in the game's data handling practices, specifically regarding how user voice communications are utilized for Artificial Intelligence (AI) development. A clause in the privacy policy, which explicitly stated that voice data could be processed to train AI models, has been quietly removed following community scrutiny. This discovery highlights the growing intersection between gaming, user privacy, and the rapid integration of generative AI technologies by major developers like HoYoverse.
The Controversial Clause and Its Removal
The issue was first brought to light by a user named Cevian on the social platform Bluesky, who spotted a specific subsection in the game's privacy policy published in January 2026. The now-deleted text, identified as Section 1, subsection xviii, outlined the collection and processing of data from voice-based chat functions.
Prior to its removal, the policy stated: "Where applicable, our services may offer voice-based chat functions that allow users to communicate with others in real time using voice. If you choose to use such features, we may process voice communication data for the purposes described in section 2 and section 13 of this Privacy Policy." The text further clarified that this processing included providing communication services, security, content moderation, and compliance with regulations.
While the specific subsection regarding voice data has been scrubbed from the document, Section 2 of the policy remains active and contains broad language regarding AI training. It states: "Like many AI-powered experiences, we may use your user-generated input such as chat data to train and improve the model that we use to provide our Services." This section assures players that they can opt-out of model training via their service-related settings without affecting gameplay.
HoYoverse's Push into Generative AI
This privacy policy adjustment occurs against the backdrop of HoYoverse's increasing investment in AI technologies. The company has previously been transparent about embracing generative AI, releasing a teaser late last year for an upcoming Massively Multiplayer Online (MMO) project built on Unreal Engine 5.
This unreleased project is set to feature an "Artificial Intelligence Stimulated Eco System," "All-Terrain Battle AI," and "Intellectual NPC" mechanics. The integration of such advanced features requires vast amounts of training data, which often leads developers to leverage user-generated content. However, the silent removal of the voice-specific clause suggests a potential pivot in strategy or a response to privacy concerns, although HoYoverse has not yet issued an official comment clarifying why the subsection was deleted.
My Take
The silent removal of this clause is a double-edged sword. On one hand, it shows that HoYoverse is reactive to community scrutiny regarding privacy. On the other, the lack of a formal announcement regarding the change creates ambiguity. As gaming companies race to build "living" worlds with generative AI, the line between gameplay data and biometric data (like voice) becomes critical. Players need explicit, opt-in mechanisms for biometric data training, not just buried clauses in a sprawling privacy policy. Until HoYoverse clarifies whether voice data is still considered "user-generated input" under Section 2, users concerned about privacy should proactively check their in-game settings to opt out of model training.