Incogni’s Gen AI and LLM Data Privacy Ranking 2026 evaluates 13 AI platforms across three privacy dimensions—data handling, transparency, and personal data collection—to gauge how they manage user information.

In the near term, concrete triggers before re-signing in include Claude’s opt-out deadline and Copilot’s data-broker practices.

The piece notes related disclosure context by mentioning Ziff Davis, the parent company of ZDNet, is involved in a separate OpenAI-related litigation.

Anthropic’s consumer policy now retains training-enabled chats and coding sessions for five years, with a 30-day opt-out retention window; enterprise terms differ and are not covered by these consumer changes.

Google Gemini is criticized for relying on a partial privacy policy that defaults to Google’s general terms, while xAI’s Grok trains on both user data and public X conversations.

Incogni provides practical guidance: treat inputs as shared with a third party, avoid sensitive data, prefer web-based versions for privacy, and consider on-device or local models for highly sensitive tasks.

Lower privacy-risk scores correlate with restricted third-party data sharing and minimal app data collection across leading platforms.

No platform offers full data privacy or retroactive data withdrawal for training; rankings measure transparency and sharing practices rather than a binary safe/unsafe verdict.

The study identifies Microsoft Copilot, Meta AI, and Moonshot AI’s Kimi as the highest privacy-risk platforms, with Kimi highlighted for its handling of user data and opt-out limitations.

Inflection AI’s Pi earns the lowest overall privacy-risk score, while Mistral AI’s Vibe leads in the category of “what happens to user data,” with Z.ai ranking among the top performers.

Anthropic updated Claude’s policy (effective July 2026) to require active opt-out to prevent inputs/outputs from training consumer accounts, with retention differing between opt-in and opt-out and some training still possible in safety reviews.