Data protection scrutinized: Study evaluates and ranks AI models on user privacy – Notebookcheck
The use of artificial intelligence (AI) in everyday life is increasing, and with it the importance of data protection. Data protection analysts at Incogni recently evaluated nine leading Large Language Models (LLMs) to determine which platforms most respect user privacy. The study evaluates the AI models based on eleven criteria divided into three main categories: data collection for training, transparency and data use and sharing.
Le Chat from Mistral AI leads the ranking. According to Incogni, the model collects comparatively little user data and offers a very data-efficient approach, both during the training phase and in live operation. Following closely behind is ChatGPT from OpenAI, which primarily scores points for its transparency. Users can specifically specify in their account whether their chats can be used to further develop the model, and if history is deactivated, storage is prevented. Grok from xAI takes third place, also impressing with its transparent data protection practices.
At the bottom of the rankings are Meta AI, Google’s Gemini and Microsoft’s Copilot. According to Incogni, these platforms collect extensive data without offering users clear opt-out options. DeepSeek, a Chinese model, also performs poorly. These providers do not offer clear options for excluding user data from training.
Incogni analysts evaluated the platforms based on eleven criteria divided into three categories: data collection for training, transparency, and data use and sharing. The "data use and sharing" category accounted for 50% of the overall rating, while "transparency" accounted for 30% and "training data" for 20%. Only four of the nine tested tools allow users to actively exclude their data from training.
The study shows that data protection is not yet standard everywhere in the world of AI. Users should therefore actively familiarize themselves with the providers’ privacy policies, especially when it comes to sensitive content, confidential customer information or personal data. Those who value data protection will find reasonable options in Le Chat, ChatGPT or Grok. However, truly data-efficient LLMs will still be the exception in 2025.
Incogni
