This seems like resurrecting a dead horse, but it’s just statistics that the puff article about AI and privacy has some relevancy in these times and can act as a thread for more interesting investigations into the matter. From PCMag:
ChatGPT, Copilot, DeepSeek, or Gemini: Which AI Chatbot Collects the Least of Your Data?
I scanned the privacy policies and reports for the top AI chatbots and found that just one doesn’t share your data with advertisers or use it for training purposes. For now, anyway.
[…]
One app report is significantly longer than the others, and anyone who has read my recent article on invasive apps could likely guess its parent company. Yep, it’s Google.
Gemini’s reported data collection includes your browsing history, contact list, emails, photos, precise location, search history, texts, and videos. It seems a bit much, especially compared with Qwen, which claims to collect little more than your device ID and app interactions.
However, Qwen’s privacy report doesn’t quite match the claims made in its privacy policy. It’s worth noting that the privacy reports in Apple’s App Store and Google’s Play Store contain information that is self-reported by the companies. The reports are not verified by independent organizations or even the app stores themselves.
[…]
At a glance, most apps collect a fairly large amount of customer data, with one exception: Microsoft’s Copilot. Microsoft’s privacy documents require bouncing through a few links, as Microsoft 365 Copilot has different data protection than Bing, which is used for search in the app. Copilot’s privacy policy states that it collects minimal customer data, doesn’t send your info to advertisers, and refrains from using your prompts and responses to train foundation LLMs, which is ideal.
That said, it’s important to note that Microsoft has a partnership with OpenAI. Both Copilot and ChatGPT use OpenAI’s GPT, though it’s implemented differently. Copilot is integrated into Microsoft 365 and utilizes your own data to generate contextual answers tailored to your work. Meanwhile, ChatGPT models use and contribute to a much larger public dataset. When you enter prompts or other data into ChatGPT, that data is pulled into the public dataset and used to train the LLM. With Copilot, this is not the case, which is how Copilot Studio meets various data protection standards, including FedRAMP, HIPAA, and SOC.
[NOTE: GDPR is not mentioned, so I ques the EU regulatory requirements about data slurping do not apply in this articles comparisons]
[…]
How to Stop AI Chatbots From Collecting Your Data
Avoid using AI chatbot apps on your phone. Instead, you can access them by running the models locally on your computer without giving up so much data.
This all avoids the real issue. Someone reveals all this data to the service via the actual chat contents. No matter what promises are made to not train or save or log the data, this has often been revealed as false. And even if true.. the data HAS to make it to some system to be processed somehow to generate responses.
No amount of using an app vs not or privacy policies or anything else can get around that.
Or just avoid using chat bots in general. In 99% of cases they have nothing to say which I cannot find myself in the same or less amount of time (because I have to verify information anyway).