Meta's AI Incognito Chat hides conversations but not screenshots
Meta's AI Incognito Chat promises private AI chats—but can't stop screenshots
Meta has launched AI Incognito Chat, a feature designed to process AI conversations in a secure, isolated environment using its Private Processing system. The system leverages trusted execution environments (TEEs) to encrypt requests between devices and processing units, hiding user IPs through third-party routing. Conversations are automatically deleted after sessions end, making it suitable for sensitive queries.
However, the feature’s transparency mechanism—hosted by Cloudflare—has faced scrutiny. A security firm found that 30 of 40 examined namespaces were stuck at their initial state, raising doubts about verifiability. While Incognito Chat prevents Meta from accessing prompts or responses, users can still copy, screenshot, or save outputs, and the system lacks explicit enterprise controls like audit logging or DLP integration. Businesses handling regulated data should verify these gaps before adoption.
- Meta AI Incognito Chat uses server-side Private Processing with TEEs to encrypt and isolate AI conversations, hiding them from Meta’s servers.
- Conversations auto-delete after sessions end, but users can still screenshot outputs, and the system’s verification layer remains unverified.
- No explicit enterprise controls (audit logs, DLP) or model-training policies are specified, requiring businesses to seek written confirmation.
Why It Matters
Professionals handling sensitive data gain a privacy tool, but must verify gaps before trusting it for regulated use cases.