Guide from the official docs

Claude Code sent users' email addresses to websites: what happened, how to check, what to change

Claude Code put signed-in users' account email in User-Agent headers. What issue #107757 and Anthropic say, which versions have the privacy mod, and how to check yours.

By AideDex desk · published

This guide follows the official docs and announcements, linked inline. AideDex did not test these products and does not score them. How we check facts.

Last checked Nov 6, 2026, 06:15 UTC. Times below are UTC.

When you are signed in to Claude Code, the model sees your account email address, and in some sessions it wrote that address into the User-Agent header of curl commands it ran, so the sites it fetched received it. A user reported this on GitHub on Nov 5 (issue #107757). Anthropic’s replies say version 2.1.317 and later ship a built-in check called cc-plugin-privacy, and that a release removing the email from the model’s context should arrive “early next week” (Hacker News). We have not reproduced the bug ourselves; the checks below follow the issue thread, Anthropic’s release notes and the Claude Code docs.

What the bug report says

Issue #107757, titled “Bug Report: Privacy violation in User-Agent header”, was opened at 22:13 on Nov 5 by GitHub user taylor-edwards, on Claude Code 2.1.318 on macOS (GitHub). The reporter writes that Claude Code “divulged my email to arbitrary 3rd parties in the User-Agent request header while conducting web searches without my permission”, and links that to “dozens of scam emails per day”. They ask for the behavior to be removed from the harness rather than made a setting. The issue carries the labels bug, platform:macos, area:tools and area:security, and it was still open when we read it.

Four minutes later it was posted to Hacker News as “Claude Code sends your email address to websites” (Hacker News), and that thread is where most of Anthropic’s explanation appeared.

Why the model had your email

A Claude Code team member posting as lydia_hallie gave Anthropic’s account at 04:24 on Nov 6 (Hacker News), opening with “I work on Claude Code”; a commenter copied the post into the GitHub thread (GitHub). Users “had reported Claude would hallucinate names for things like @author tags”, the post says, so Anthropic “added the account email in the model’s context for signed-in users”. When Claude wrote its own command, such as a curl call, “it sometimes put your email address in the User-Agent as a contact address”, because “some sites ask agents for a contact”. It ends with “this should not have happened, and we should’ve caught this behavior earlier.”

The email in the context was reported long before. Issue #81138, opened Jul 25, 2026, says the account email is added to the system prompt as “The user’s email address is …” and is read from oauthAccount.emailAddress in ~/.claude.json; it was still open on Nov 6. On Aug 17, release 2.1.234 added a softer guard: “Claude is now told to use your account email only to identify you, and not to send it to unrelated services unless you ask” (v2.1.234 release, Changelog). That is an instruction to the model, and the Nov 5 report came about 11 weeks after it.

Date (UTC)What happenedSource
Jul 25, 2026Issue #81138 reports the account email in the system promptGitHub
Aug 17, 2026v2.1.234 tells Claude to use the email only to identify youRelease, Changelog
Nov 3, 2026v2.1.317 adds a check before a tool call sends your account emailRelease, Changelog
Nov 4, 2026v2.1.318, the reporter’s version; its notes don’t mention emailRelease, Changelog
Nov 5, 22:13Issue #107757 openedGitHub
Nov 5, 23:26 and 23:31poteat on GitHub and mpoteat on Hacker News, writing as the Claude Code team (“we introduced a built-in mod”), points to cc-plugin-privacyHacker News, GitHub
Nov 6, 00:00Source of the privacy mod published in the repoGitHub
Nov 6, 04:24lydia_hallie: email to be removed from context in the next releaseHacker News

On Hacker News, mpoteat wrote that Anthropic “were initially aware of this a few days ago when this problem started popping off on Reddit” (Hacker News).

Who is affected

By Anthropic’s account, the email went into the context “for signed-in users” (Hacker News), and #81138 traces it to the signed-in account record (GitHub). In that account, the address left the machine when Claude wrote it into a command. Neither Anthropic reply we read gives a count of affected users or names the sites that received addresses.

What you have depends on the version:

  • Before 2.1.317: no built-in check. The v2.1.317 notes are the first to add one: “a check that asks before a web or shell tool call sends your account email address to the network” (v2.1.317 release, Changelog).
  • The stable channel: on Nov 6 the npm stable tag pointed to 2.1.311 and latest to 2.1.318 (npm registry). The lydia_hallie post says the same: “The stable channel is still on 2.1.311, so update to 2.1.317 or later” (Hacker News).
  • 2.1.317 and 2.1.318: the mod is meant to be on by default (GitHub, Hacker News). Two commenters on 2.1.318 wrote that they could not find cc-plugin-privacy in their plugin list (francescor, BigBalli). One commenter on 2.1.318 reported the opposite: cc-plugin-privacy showed as enabled and built-in in /plugins, yet a test command holding their account address next to User-Agent ran without a refusal (love540). We have not tried to reproduce it, and the thread had no Anthropic answer to it when we read it.

What the privacy mod checks, and what it lets through

Anthropic describes the mod in three places, and they don’t match exactly. The release note says it “asks before a web or shell tool call” sends the address (v2.1.317). The lydia_hallie post says it “asks you before Claude runs a web tool call or a curl-style command containing your email” (Hacker News). The README published with its source is narrower (GitHub):

  • It refuses a Bash, PowerShell or Monitor call whose input contains your account address and names a User-Agent (User-Agent, --user-agent, curl -A, wget -U and similar). The refusal “raises no prompt”, and a saved allow rule or auto mode does not clear it.
  • “WebFetch and WebSearch are not checked, and neither are MCP tools.”
  • The address with no User-Agent named passes: in a URL, a request body, curl -u or a From: header.
  • A value worked out at run time, such as $(git config user.email), a variable, a file, or a script that holds the address also passes.

The README calls the mod “a courtesy, not a boundary”. The lydia_hallie post agrees on scripts: the mod “still doesn’t prevent all cases, like when Claude writes a script and runs on its own” (Hacker News).

How to check your setup

  1. Version. Run claude --version in a terminal; the setup docs use the same command to confirm an install. You want 2.1.317 or later.
  2. The mod. In a session, run /plugin, press Tab to reach Installed, and look under Built-in, where the docs say Claude Code lists its own mods (Mods overview, Discover plugins). The entry is cc-plugin-privacy (GitHub). The docs’ table of built-in mods did not include it yet when we read it on Nov 6.
  3. Nothing has switched it off. The README lists what disables it: /plugin disable, "cc-plugin-privacy@builtin": false under enabledPlugins in user, --settings or managed settings, a hook of another installed plugin that overrules it, or a server-side switch at Anthropic (GitHub). A repository’s settings can’t turn it off. Your user settings live in ~/.claude/settings.json (Discover plugins).
  4. Which address. The email Claude Code holds for you shows in /status; the docs mention it when describing IS_DEMO, which hides it there (Environment variables). Two reports in the issue threads point to oauthAccount.emailAddress in ~/.claude.json as the source (#81138, #107757).
  5. Past sessions. Claude Code keeps transcripts in plain text under ~/.claude/projects/, 30 days by default (Data usage). To find sessions where your address and the word user-agent appear in the same transcript, run grep -rl "you@example.com" ~/.claude/projects/ | xargs grep -il "user-agent" with your own address. A hit tells you which transcript to open; read the command to see where it went.

How to limit it until the fix ships

  • Get onto 2.1.317 or later. claude update follows your release channel; the default "latest" gets new versions as they ship, and "stable" uses a version “typically about one week old” (Setup docs). On Nov 6, stable meant 2.1.311, which has no mod (npm registry). You switch channels with autoUpdatesChannel in settings or in /config.
  • Leave cc-plugin-privacy on, but don’t treat it as protection: the README calls it a courtesy, and one commenter reported it not refusing a test command (see above). The flags that stop installed mods, disableAllHooks, --bare and --safe-mode, don’t stop built-in mods, per the Mods overview.
  • Deny the network commands. The permissions docs suggest deny rules for curl, wget and similar, with WebFetch allowed per domain, and warn that a deny rule “doesn’t match the same program by path or inside sh -c”, so they pair it with the sandbox network allowlist (Configure permissions):
{
  "permissions": {
    "deny": ["Bash(curl *)", "Bash(wget *)"]
  }
}
  • Expect little from a CLAUDE.md line. You can tell Claude never to put your address in a header, but the same docs page says CLAUDE.md guidance “shapes what Claude tries but doesn’t enforce a boundary” (Configure permissions).

We found no documented setting that keeps the address out of the model’s context. The environment variables and settings references we read on Nov 6 list none (Environment variables, Settings), and #81138 said the same in July (GitHub).

When the fix ships

Anthropic’s stated fix is to stop giving the model your email: “We’re removing the account email address from the context Claude Code gives the model. This update will hit Claude Code users in the next release, which should arrive early next week” (Hacker News, quoted in GitHub). The post promises an update in the Hacker News thread once it is out. At 06:15 UTC on Nov 6 the newest release on GitHub and in the docs changelog was 2.1.318, and neither mentions the change (GitHub releases, Changelog).

For what agents keep about you in general, see AI agent memory and Is it safe to connect Gmail to an AI agent?. Another vendor’s privacy reports are collected in Meta Muse privacy incidents, and agents that touch accounts are covered in Can an AI agent log into your bank?. Anthropic’s own agents have pages too: Claude Cowork safety, Claude in Chrome, and Claude Duet vs Hermes Agent for its personal agent set against an open-source one.

Organizations that need to control Claude Code more broadly can read what administrators can lock, and what they cannot, in Claude Code managed settings.

FAQ

Why would a script put an email address in a User-Agent at all?

Some sites ask for it. Wikimedia’s policy says “Scripts should use an informative User-Agent string with contact information, or they may be blocked without notice”, and lists an email address, a website or a wiki username as acceptable contact details (Wikimedia Foundation). The MediaWiki API etiquette page gives the format clientname/version (contact information e.g. username, email) (MediaWiki). A username or project page satisfies both, so the address is not required.

Doesn’t HTTP have a header meant for the user’s email?

Yes, From. The HTTP standard says a user agent “SHOULD NOT send a From header field without explicit configuration by the user, since that might conflict with the user’s privacy interests”, while a robotic user agent should send one so its operator can be reached (RFC 9110, HTTP Working Group). The privacy mod’s README lists an address in a From: header among the cases it lets through.

Does setting IS_DEMO stop the model from seeing my email?

The docs describe IS_DEMO as hiding your email and organization name from the header and /status output, and say nothing about the model’s context (Environment variables). Note that setting it to 0 or false still turns demo mode on; you have to unset it.

Can I read the privacy mod’s code?

Yes. Anthropic added it to the mods/privacy folder of the Claude Code repository in commit 90380ef at 00:00 UTC on Nov 6, about half an hour after poteat wrote that they would try to publish it (GitHub, issue comment). The check lives in hooks/register.ts.

Sources

13 pages
  1. GitHub: anthropics/claude-code issue #107757, Privacy violation in User-Agent header (Nov 5, 2026) github.com
  2. Hacker News: Claude Code sends your email address to websites news.ycombinator.com
  3. Hacker News: comment by lydia_hallie, Claude Code team (Nov 6, 2026) news.ycombinator.com
  4. GitHub: Claude Code v2.1.317 release notes (Nov 3, 2026) github.com
  5. Claude Code Docs: Changelog code.claude.com
  6. GitHub: claude-code mods/privacy README github.com
  7. GitHub: anthropics/claude-code issue #81138, account email injected into the system prompt (Jul 25, 2026) github.com
  8. Claude Code Docs: Mods overview code.claude.com
  9. Claude Code Docs: Configure permissions code.claude.com
  10. Claude Code Docs: Set up Claude Code (release channels) code.claude.com
  11. npm registry: @anthropic-ai/claude-code dist-tags registry.npmjs.org
  12. Wikimedia Foundation User-Agent Policy foundation.wikimedia.org
  13. RFC 9110, HTTP Semantics: From header field rfc-editor.org

Spotted something wrong or out of date? Report an error · Corrections log