The privacy paradox of AI: New ranking finds the most popular chatbots may pose the greatest data risks

Generative artificial intelligence has become embedded in everyday life. Millions of people now use chatbots to draft emails, analyse documents, brainstorm ideas, discuss personal concerns, and even seek professional guidance. As AI adoption accelerates, a critical question is emerging: what happens to the information users share with these systems? A new study from data privacy company Incogni suggests that consumers should pay closer attention to the privacy practices of AI platforms. The company’s 2026 Gen AI and LLM Data Privacy Ranking assessed 13 leading chatbot services and found that some of the world’s most widely used AI tools pose among the highest privacy risks to users.

The research evaluated ChatGPT, Claude, Gemini, Grok, Vibe, Perplexity, Qwen, DeepSeek, Z.ai, Kimi, Meta AI, Pi, and Copilot across 11 privacy-related criteria spanning data usage, transparency, and data collection practices. Lower scores reflected less privacy-invasive approaches. The study’s main finding is that size and popularity do not necessarily correlate with stronger privacy protections.

According to Incogni’s ranking, Microsoft Copilot, Meta AI, and Google’s Gemini received some of the highest privacy-risk scores among the platforms assessed. By contrast, Mistral AI’s Vibe (formerly Le Chat) achieved the lowest risk score and was ranked as the most privacy-friendly platform. OpenAI’s ChatGPT and Inflection’s Pi also performed relatively well.

The results challenge a common assumption that the largest technology companies automatically provide the strongest privacy safeguards. Instead, Incogni’s analysis suggests that users may need to examine individual privacy policies and data practices rather than relying on brand recognition. Privacy has increasingly become a competitive issue within the AI sector. While model performance, reasoning capability, and speed remain important differentiators, users are beginning to ask how their conversations are stored, analysed, and potentially reused.

ChatGPT scores highest for transparency

One area where OpenAI stood out was transparency. The study identified ChatGPT as the platform with the clearest and most accessible privacy documentation. Researchers found that OpenAI provided comparatively straightforward explanations about how user information is handled and whether conversations may be used for model training. Transparency is becoming increasingly important because AI privacy policies are often complex and difficult for non-specialists to understand. Users may agree to terms and conditions without appreciating how their data might be processed, retained, or shared. Incogni’s assessment suggests that clear communication about data usage may be just as important as the underlying privacy controls themselves.

The most concerning finding from the report is what remains unknown. Despite extensive public debate about AI training practices, Incogni found that none of the assessed providers disclosed the precise datasets used to train their models. Instead, companies typically use broad descriptions such as “publicly available information,” “social media data,” or “private datasets.” This lack of specificity makes it difficult for individuals to determine whether personal information collected elsewhere online may have contributed to model development.

The issue has attracted growing regulatory scrutiny. Legislators and regulators across North America and Europe are increasingly focused on questions surrounding data provenance, consent, copyright, and the use of publicly accessible information in AI training systems. For consumers, however, the practical reality is that there remains limited visibility into exactly how many AI models have been trained and what information may have been incorporated during the process.

Can users remove their data?

The report delivers a key conclusion for privacy advocates. Incogni found that no platform currently allows users to remove information that has already been incorporated into model training. While many providers offer mechanisms allowing users to opt out of future training use, these controls do not operate retrospectively. Once information has been used for training, there is currently no practical mechanism for reversing that process. This distinction is important because many users assume deleting a conversation or opting out later removes all traces of their information. According to the research, that assumption may be incorrect.

Cybersecurity expert Miguel Fornés of Incogni argues that users often treat AI systems as trusted confidants despite the uncertainty surrounding data retention. In comments accompanying the study, Fornés warned that individuals frequently share highly sensitive personal or commercial information with chatbots, potentially without understanding how that information may be used in the future.

Opt-out options exist, but not always prominently

The study did identify some encouraging developments. According to Incogni, nine of the thirteen platforms assessed provide users with a relatively straightforward mechanism, often a settings toggle, to prevent future conversations from being used in AI training. These include ChatGPT, Claude, Gemini, Copilot, Grok, DeepSeek, Perplexity, Pi, and Vibe. However, the usability of these controls varies considerably. Some providers place privacy settings prominently within user accounts, while others require users to submit forms, send emails, or navigate complex account-management processes.

This creates a familiar challenge in digital privacy: protections may technically exist, but users cannot benefit from them if they are difficult to locate or understand.