Blog · AI Agent News · Safety & policy VIEN
Claude Incognito Chats Still Know Who You Are: Here Is Why
Mathematician David Bessis caught Claude Opus 5.5 naming him inside incognito thinking, exposing what incognito mode actually hides and keeps.

In this article
On November 16, 2026, mathematician David Bessis posted on X that he opened Claude Opus 5.5 in incognito mode just to proofread text, only to see his own name appear inside the model's thinking output. His main post reached 1,768 likes, 141 replies, and 313,200 views, while a follow-up confirming incognito mode gathered 183 likes and 30,432 views, prompting a closer look at how AI private modes work under the hood.

Why did Claude incognito mode still know his name?
Spotting his own name inside an unprompted proofreading chat led Bessis to call the behavior a "distressing level of dishonesty." When pressed inside the thread, Claude replied that the user's account email traveled with each request as background context.
It helps to separate Anthropic's two products here: on Claude's web app, incognito stays detached from chat history and memory while remaining linked to a signed-in account; in the Claude Code CLI, Hacker News item 50285046 showed account emails still sitting inside model context, with a community member (AlexErrant) publishing an unofficial plugin to strip it, rather than an Anthropic-shipped fix. Anthropic has not issued a separate public statement on Bessis's X post, yet Claude's official support page already clarifies that incognito does not mean full anonymity.
flowchart LR A[Start incognito chat]:::data --> B[Skip history and Memory] B --> C[Profile details still load]:::human C --> D[Retained on servers 30 days]:::key classDef key fill:#1d1a10,stroke:#ffd23f,color:#ffd23f,stroke-width:1.6px classDef human fill:#181817,stroke:#e8e6df,color:#e8e6df,stroke-dasharray:5 3 classDef data fill:#1c1c1b,stroke:#8d8b85,color:#cfcdc6 classDef muted fill:#181817,stroke:#3d3d3a,color:#8d8b85
What does Claude incognito mode actually hide?
Available across Free, Pro, Max, Ultra, Team, and Enterprise plans for conversations started outside of a Project, incognito mode opens via the ghost icon with a black border. Claude Help Center documents the exact boundaries below:
| Data aspect | Included or saved in incognito? | Official Claude Help Center detail |
|---|---|---|
| Chat history & past-chat search | Not saved | Sessions cannot be reopened or converted once closed |
| Claude Memory & monthly recap | Not read or written | Excluded from existing and future memory entries |
| Model training | Not used | Applies across consumer and commercial plans |
| Profile info (custom styles, preferences) | Accessible | Official FAQ states Claude can access profile details |
| Back-end server retention | Kept 30 days by default | Enterprise orgs may configure longer custom retention |
| Team & Enterprise admin exports | Included | Visible in owner exports and Enterprise Compliance API |
Also readGemini 4 Hydrogen Review: Fast & Cheap, Yet Weak on 3DHow should users protect sensitive inputs?
Because your session remains tied to an authenticated account and retained for 30 days, incognito behaves far more like "keep off the sidebar" than a zero-knowledge vault. Build four habits before pasting sensitive material into any cloud chatbot:
- Redact identifiers first: replace real names, email addresses, ID numbers, and client names with generic placeholders before sending.
- Audit your profile settings: clear personal details stored in custom styles or account preferences since incognito sessions still read them.
- Never mix personal files into work accounts: avoid pasting private documents into Team or Enterprise seats where admins can export logs.
- Keep strictly confidential work local: run local open-weight models on your own hardware when documents cannot leave your machine.

Where I stand
A ghost icon and a black border naturally invite users to assume a session vanishes the moment the tab closes, yet it only skips sidebar history and model training. Spending 30 seconds scrubbing names, emails, and client records locally before hitting send beats trusting any privacy toggle on a web UI.
Sources
- David Bessis on Claude incognito chat X (@davidbessis)
- David Bessis clarifying incognito mode X (@davidbessis)
- Use incognito chats Claude Help Center
- While we're modding Claude Code, Anthropic includes your email in the system prompt Hacker News
Safety & policyClaude & Anthropic
Agent basics · Nov 17, 2026Caltech Professor Stops Grading Math Homework Accuracy Over AIAfter 17 years at Caltech, math professor Yi Ni now grades homework on completion only because chatbots solve college sets, raising warnings…
Safety & policy · Nov 17, 2026California Sues OpenAI After 80+ Agent Control FailuresCalifornia AG Rob Bonta sues OpenAI over pre-release agent breaches including Hugging Face, offering a three-layer guardrail lesson for AI agent builders.
Coding agents · Nov 17, 2026Claude Code Mode vs. Tokendoctor: Cut Token Spend Up to 38%?Comparing Claude Code Mode JS tool batching and Tokendoctor context tuning: real numbers, exact setup steps, strict limits, and who benefits.