Google Now Uses Your Photos and Voice Recordings to Train Its AI
Google has updated its privacy settings to include images, audio, and video in its AI training data. Users are opted in by default and must manually adjust settings to opt out.
Google has quietly expanded what it collects from users to train its artificial intelligence models. The update allows the company to retain images, files, and audio or video recordings submitted through its services. As web-scraped data faces legal and ethical scrutiny, Google is turning to user-generated content to keep its generative AI development moving.
The change arrives through two new settings: Search Services History and Personalized Recommendations. These controls govern how activity across Maps, Shopping, Flights, and Translate is stored and used. A photo sent through Google Lens or a voice clip processed by Google Translate may now be retained for research and used to develop future technologies. Google has confirmed that this data feeds generative model training and is occasionally reviewed by human staff for safety and accuracy.
Before this update, most data retention fell under the Web and App Activity setting. By carving out Search Services as its own category and enabling it by default, Google effectively resets the preferences of users who had previously restricted their data sharing. To maintain those restrictions, they must now navigate new menus. Options include unchecking the Save Media box entirely or setting an auto-delete schedule between three and 36 months.
Google is not alone in this approach. Meta recently drew criticism for using camera roll images and footage from its smart glasses to train internal models. The pattern is consistent across major platforms: large proprietary datasets are now treated as a competitive asset, and the burden of opting out falls on individual users rather than being offered as a clear choice from the start.
What This Means for African Users
For users across Africa, this update carries specific concerns that go beyond the general privacy debate. Millions of people in Nigeria, Kenya, South Africa, and across the continent depend on Google Translate and Voice Search to navigate local languages and dialects. Google positions this data collection as a path to better AI, but there is little transparency on whether African users will see direct improvements in local language support as a result. The benefit flows to Google's models; the return to users is unclear.
The policy also has potential regulatory consequences. Nigeria's Data Protection Act and similar frameworks being developed in other African countries place specific obligations on how companies collect and process personal data. Default opt-in practices for AI training could draw regulatory attention and put Google in a difficult position as enforcement capacity on the continent grows.
There is a more immediate practical issue as well. Users on limited mobile data plans, which remain the norm across much of sub-Saharan Africa, may find that background uploads of high-resolution media consume paid bandwidth without their active awareness. That is a real cost, not just a policy concern.
Transparency in data collection is not just a principle; for many African users, it has a direct price attached to it.
