A new report from Incogni evaluates the privacy practices of 13 major AI platforms, identifying significant differences in how user data is collected, shared, and used for model training.
Key facts
- •The study evaluated 13 platforms, including ChatGPT, Claude, Gemini, Grok, Vibe, Perplexity, Qwen, DeepSeek, Z.ai, Kimi, Meta AI, Pi, and Copilot.
- •Lower risk scores were assigned to platforms with more transparent data-sharing practices and clearer privacy policies.
- •No platform currently allows users to retrieve their data once it has been used to train a large language model.
- •Meta was characterized as the most data-hungry model in the sample, with policies that are difficult to verify.
- •DeepSeek provides users the ability to request the correction or removal of inaccurate personal information.
A study titled "Gen AI and LLM Data Privacy Ranking 2026" by Incogni has evaluated 13 AI platforms based on their privacy risks. The research analyzed how these services handle user conversations, data sharing with third parties, and transparency in their privacy policies. The findings suggest that while some platforms offer clearer data practices, others present higher risks regarding user information and model training.
Top and Bottom Performers
According to the report, Vibe, ChatGPT, and Pi achieved some of the lowest risk scores, indicating better privacy practices. Conversely, Gemini and Meta AI were among those receiving the highest risk scores. Incogni identified Moonshot AI's Kimi as representing the greatest privacy risk among the platforms assessed, citing difficulties in opting out of model training and the app's use of data to track users.
Data Handling and Transparency
The study highlights varying levels of transparency across platforms. For instance, Mistral AI's Vibe is noted for its privacy-friendly mobile app and minimal third-party data sharing. In contrast, Microsoft's Copilot was cited for sharing data with third-party advertisers and purchasing data from brokers. Other platforms like Claude have shifted policies, with the July 2026 update requiring users to opt out of data usage for AI training.
Recommendations for Users
Incogni advises users to treat any information entered into an LLM as data provided to a third-party service. Miguel Fornés, an information security manager at Incogni, recommends avoiding the input of passwords, financial details, medical records, or confidential work information. Users are encouraged to prioritize platforms with transparent policies and consider using locally run models for sensitive tasks.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by ZDNET AI.
