Data Protection in LLMs ChatGPT, Gemini, And Others Under Scrutiny: Which AI Snoops the Most?

Related Vendors

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

Wuxi InfiMotion Technology Co., Ltd.

The privacy analysts at Incogni have examined well-known large language models based on various criteria to determine how secure your private data is. In the ranking, Le Chat, ChatGPT, and Grok lead as they intrude the least into users' privacy. Gemini, DeepSeek, and Meta AI, however, snoop extensively.

Artificial intelligence has become part of everyday life for most people on the planet, often so subtly that many don't even notice it. Sure, voice assistants, text generators, or research helpers are consciously utilized by users. However, AI searches in browsers or apps are often less noticeable. The more naturally tools like ChatGPT, Copilot, or Gemini are interacted with, the more urgent the question becomes: What happens to users' data?

The privacy-focused company Incogni has examined nine leading Large Language Models (LLMs) as part of the Privacy Ranking Report 2025. These include ChatGPT (OpenAI), Gemini (Google), Copilot (Microsoft), Meta AI, Grok (xAI), Le Chat (Mistral AI), DeepSeek, Claude (Anthropic), and Pi (Inflection AI). The evaluation was based on a proprietary privacy ranking that encompasses eleven criteria, organized into three main categories: data collection for training, transparency, and data use and sharing.

Gallery

Gallery with 8 images

Clear Winners: Le Chat, ChatGPT, And Grok

The best performer is "Le Chat" from Mistral AI. According to Incogni, the model collects comparatively little user data and provides a highly data-efficient foundation, both during the training phase and in live operation. Close behind is ChatGPT from OpenAI. This may surprise many early AI critics, but the service mainly scores with transparency.

Users can specifically determine in their account whether their chats may be used for further development of the model, and when history is disabled, storage is prevented. This is by no means the case with every LLM. Third place is taken by Grok from xAI (Elon Musk's AI venture), which also impresses with transparent data protection practices.

Trailing Behind: Gemini, Copilot, Meta AI

The results are less favorable for other prominent providers. Meta AI surprisingly ranks last in the ranking, even behind the Chinese model DeepSeek. According to Incogni, Meta AI extensively collects data without providing users with clear opt-out options. Google Gemini and Microsoft Copilot also rank far behind, largely because, as noted in the Incogni study, they draw training data from user inputs and provide little proactive information about its usage.

Method With Depth

The analysts at Incogni did not weight the criteria equally: the category "Data Use and Sharing," which addresses what happens to personal information once it is collected, accounted for 50 percent of the overall rating. The categories "Transparency" (30 percent) and "Training Data" (20 percent) rounded out the evaluation. Only four of the nine tested tools allow users to actively exclude their data from training, including ChatGPT, Grok, Le Chat, and Copilot. For Gemini, DeepSeek, Pi AI, and Meta AI, this is not possible. For Claude, it is stated that user data has never been used for training the models.

The ranking clearly shows that data protection in the world of artificial intelligence is far from being standard. There is a vast difference between a clear opt-out and a patchy privacy policy, even among big tech providers. The analysts at Incogni assume that the issue partly lies in the fact that existing company privacy policies are being applied to AI services, rather than being specifically designed and offered for these services.

Companies and individuals looking to use AI tools should actively review the providers' privacy policies, especially when dealing with sensitive content, confidential customer information, or personal data. Those who want to take data protection seriously based on the Incogni results will find much clearer and traceable options with Le Chat, ChatGPT, or Grok than with Gemini or Copilot. However, it is also clear that truly data-efficient LLMs are still the exception in 2025. (sb)

Secure and Compliant Authentication in Laboratories

Secure & Contactless User Authentication at GMP/GLP