The Claude Code Update You Need Is 2.1.317. Here’s the Email Problem It Patches.
TL;DR: Claude Code had your account email address in the model’s context — and when Claude wrote its own `curl` command, it would sometimes put that address into the request as a contact. Anthropic has owned it plainly: “I’m sorry we let this happen.” The claude code update that adds a guard is 2.1.317. The catch: the stable channel is still on 2.1.311, from 26 October. So most people do not have the guard yet.
What actually happened before the claude code update
On 5 November a user opened issue #107757 on Anthropic’s own repository, titled “Bug Report: Privacy violation in User-Agent header”. It is still open, labelled `area:security` and `bug`, and carries 317 thumbs-up. It reached 996 points on Hacker News.
A member of the Claude Code team replied in that thread. Their comment opens “I work on Claude Code”, and the explanation is worth reading carefully because it is not what the headline says:
“Users had reported Claude would hallucinate names for things like @author tags, since it didn’t have context on the account. So we added the account email in the model’s context for signed-in users. But when Claude wrote a command itself, like a curl call, it sometimes put your email address in the User-Agent as a contact address. Some sites ask agents for a contact, and Claude would sometimes provide the email address in its context.”
So this was not a header Claude Code stamped on every request. Another Anthropic voice in the thread explained why Claude chose to do it at all: “Some sites like Wikipedia have a policy where scripted clients identify themselves, and there’s a standard in HTTP to have a contact email in the User-Agent for such cases. Claude apparently tends to decide that this policy is worth following, and we now disallow this programmatically.”
A commenter who checked their own traffic saw the ordinary header — `Claude-User (claude-code/2.1.318 (cli); +https://support.anthropic.com/)` — with no address in it. Another put it bluntly: Claude Code “is not sending your email address to websites as a default, or regularly.”
And one thing we are not repeating. The person who filed the report blamed it for a surge of scam email. That link is disputed in the thread and nobody has evidenced it. Treat it as unproven.
Which claude code update you actually need
Three version numbers matter for this claude code update, and they are easy to mix up.
- 2.1.311 — dated 26 October, and what the team member says the stable channel is still on. No guard.
- 2.1.317 — dated 3 November. Anthropic’s own changelog describes it exactly: “Added a check that asks before a web or shell tool call sends your account email address to the network.” This is the claude code update to get today.
- The real fix — the claude code update that removes the account email from the model’s context altogether — is not shipped yet. The team member says it arrives “in the next release, which should arrive early next week”. The newest version on the changelog as we write is 2.1.318, from 4 November, and it carries no such entry.
Note what this claude code update does: it asks, it does not remove. And Anthropic is candid that it is partial — “it still doesn’t prevent all cases, like when Claude writes a script and runs on its own.” One user testing on 2.1.318 reported the guard did not block two shell commands containing their address; that is a single unverified test, but it matches what Anthropic already said.
Why this matters if you don’t code
The interesting part is not the leak. It is how it happened, which is a pattern you will meet again.
Nobody wrote code to send your email anywhere. Somebody added your email to what the model knows — for a sensible reason, so Claude would stop inventing names on `@author` tags. Then the model, left to write its own commands, used that fact the way a conscientious person might: a website asked scripted visitors to identify themselves, so it identified you.
That is the thing worth internalising. Anything in your assistant’s context is material it may use, in situations nobody anticipated. It is not a bug in the usual sense — no rule was broken. The model did something reasonable with information it should not have had.
This is a different shape of problem from the backdoored Claude Code extension last week, which was an attacker, and from shared chats turning out not to be private, which was a setting nobody understood. This one is nobody’s fault in particular and everybody’s problem in general. If “context” and “User-Agent” are unfamiliar, our vibe coding glossary explains them.
Your claude code update, step by step
- Check your version, then update. Run `claude –version`. If it starts 2.1.311 or lower, you have no guard at all. Take the claude code update to 2.1.317 or later now rather than waiting for the automatic one.
- Say no when it asks. Once the claude code update is in, you will get a prompt before a web or shell call that contains your account email. That prompt is the whole feature. Do not click through it.
- Leave the privacy mod switched on. It is on by default and can be turned off under `/plugins`. There is no good reason to.
- Watch for next week’s release. The claude code update that actually ends this is the next one — your address leaving the model’s context. Until it lands, 2.1.317 is a guard, not a cure.
- Consider a masked address for your account. Several people in the thread use one as a matter of course. It costs nothing and it limits every future version of this problem, not just this one.
Who should care (and who shouldn’t)
- Using Claude Code daily: care. Take the claude code update today, and assume your address may already be in some server logs.
- Building a revenue app: care, and check what else is in your agent’s context — this incident is a prompt to look, not just to patch.
- Using Claude in the browser or app, not the CLI: this does not touch you. It is specific to Claude Code writing shell commands.
- Website or landing page builders: not relevant unless you use the CLI.
- No idea yet: the 60-second Tool Finder quiz will tell you whether a terminal agent is even the right tool for what you are building.
Our take
Anthropic handled the disclosure about as well as it can be handled: a named engineer, in public, inside nine hours, with “Again, this should not have happened, and we should’ve caught this behavior earlier.” No deflection, a shipped mitigation, a dated fix. We would not switch tools over this.
What we would not do is treat this claude code update as the end of it. A check that asks you is a stopgap by Anthropic’s own description, and the stable channel sitting four days and six versions behind the mitigation is the part that bothers us most — the people least likely to be reading a Hacker News thread are exactly the people still on 2.1.311. If you take one thing from today, make it this: your assistant’s context is not a private notebook. It is working material, and it goes where the work goes.
Not sure a terminal coding agent is right for you at all? Take the 60-second Tool Finder quiz →
Some links on this page are affiliate links. If you buy through them we may earn a commission at no extra cost to you.
FAQ
Did Claude Code send my email to every website?
No. Anthropic’s explanation is that the address sat in the model’s context, and Claude sometimes wrote it into a User-Agent as a contact when a site asked scripted clients to identify themselves. A user who inspected their own traffic saw the standard header with no address in it.
Which claude code update fixes it?
2.1.317 adds a check that asks before a web or shell call sends your account email to the network. It is a guard, not a removal, and Anthropic says it does not catch every case. The actual fix — taking the address out of the model’s context — was said to be arriving in the following release.
Is the scam email I’m getting caused by this?
Unproven. The original reporter made that connection, and it is disputed in the thread with no evidence either way. We are reporting the leak because it is documented, and not the spam link because it is not.
