Engaging with generative AI tools raises concerns about how your inputs are used. Many AI platforms train their models using user interactions, often anonymizing data but still leaving some users uneasy. While disabling AI training is possible on some platforms, others make it difficult.
Disabling training differs from wiping chatbot history. While erasing history prevents old chats from being used for training, new inputs may still contribute. Some platforms, like OpenAI’s ChatGPT, allow users to turn off activity logging. Even with logging disabled, chats may be retained temporarily, such as for 72 hours.
Meta offers limited transparency regarding AI training. Private messages are excluded unless the AI is invited into the chat. However, public posts and images are fair game. In the UK and Europe, users can object to data collection using a “right to object” form, but in the US, users must complete a more complex process, providing detailed explanations and screenshots. Meta AI chatbots on Facebook or Instagram can be muted via settings to reduce interactions.
Other apps handle AI training differently. Adobe, for example, explicitly does not train its AI on user images, while Reddit permits OpenAI to use user posts for training without user control. Checking app privacy policies and settings is essential for understanding how your data is used.
If your content is public or accessible by third-party developers, it may still be captured by AI bots. Being cautious about what you share online is a critical step in protecting your data.
Disclaimer
NextNews strives for accurate tech news, but use it with caution - content changes often, external links may be iffy, and technical glitches happen. See full disclaimer for details.