GitHub issue #107757 (2026-11-05) reported that Claude Code put the signed-in account email into the User-Agent of web requests to third-party sites, which the reporter linked to a flood of scam emails; the issue drew 955 reactions and the HN thread hit 1,582 points. Lydia Hallie, who works on Claude Code, apologized on HN: the email was added to model context so @author tags would stop hallucinating names, and the next release (expected early next week) removes it. The interim fix is the built-in privacy mod on by default in 2.1.317+, whose own README calls it a courtesy, not a boundary.
Key Takeaways
- ✓Cause (per an Anthropic engineer): the signed-in email sat in model context, and when the model wrote its own curl it used it as a User-Agent contact
- ✓Reach: issue #107757 has 955 reactions and 103 comments; the HN thread reached 1,582 points and 506 comments
- ✓Stopgap: from 2.1.317 (npm, Nov 2) the default-on cc-plugin-privacy mod refuses Bash/PowerShell calls that hold the email and name a User-Agent
- ✓Gaps (per its README): WebFetch, WebSearch, MCP tools, scripts, variables and emails in URLs are not checked
- ✓Status: stable is still 2.1.311 and latest is 2.1.319; the real fix needs the next release that drops email from context
Key Decision Metrics at a Glance
Heavy Claude Code use: compare subscription limits and API bills
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and Industry Pain Point
Once a coding agent can write its own curl calls, any personal data in its context can leak. On November 5, issue #107757 reported that Claude Code 2.1.318 put the account email in the User-Agent of web requests to third parties, which the reporter linked to dozens of scam emails a day. The HN thread reached 1,582 points and prompted many migration and privacy complaints.
Architecture and Mechanism
Lydia Hallie, who works on Claude Code, replied on HN: to stop the model hallucinating names in @author tags, the team added the signed-in email to model context. Sites such as Wikipedia ask scripted clients for a contact, so the model sometimes filled the User-Agent with that email. She apologized and said the next release, expected early next week, removes it.
Benchmarks
There are no benchmarks for this incident. Verifiable figures: the issue has 955 reactions and 103 comments; the CHANGELOG for 2.1.317 lists a new account-email egress check; on npm, stable is still 2.1.311 and latest is 2.1.319.
Developer Guide
1) Upgrade to 2.1.317+ and confirm cc-plugin-privacy is not disabled under /plugin. 2) Read the mod README: it only refuses Bash/PowerShell calls that hold the email and name a User-Agent; WebFetch, WebSearch, MCP and scripts are not checked. 3) Enterprises can enforce it via managed settings and filter email-bearing headers at the egress proxy. 4) Sign agents in with a separate email and recheck once the real fix ships.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.