CVCOVANewsComputing, Opportunities, Value & Analysis
Cybersecurity & Privacy4 min read

Claude Code included users' email addresses in its web searches: what happened and how to reduce exposure

A report on GitHub revealed that Anthropic's coding agent could send the account's email address to third-party sites. The company rolled out a limited patch and promised a fundamental fix, which has not yet been released.

By COVA News newsroom
Close-up of a mail app icon on a phone

Photo: Brett Jordan on Unsplash

When someone asks Claude Code to look something up online, it's natural to assume that only the query goes out. One user discovered that their email address could go out too. On November 5, 2026, developer taylor-edwards opened a report, issue #107757, in the public repository for Claude Code, Anthropic's coding agent that runs commands and queries the web on behalf of whoever is using it. According to the report, the agent included his email address in the User-Agent header of the requests it made when searching the web. That header is a kind of calling card that accompanies every visit to a site, and many servers store it in their logs.

The case became one of the most discussed stories on Hacker News, the industry's go-to forum, with 1,467 points and 453 comments at the time this article was prepared. Other outlets cited lower figures, probably because they measured them earlier.

Personal data no one asked to share

The explanation came from a member of Anthropic's team, both in the issue itself and on Hacker News. According to their account, for logged-in users the account's email address was loaded into the model's context, that is, the information the AI has in view while it works. It was added because Claude was making up names when signing code as the author.

The problem surfaced somewhere else. Some sites, such as Wikipedia, ask automated programs to identify themselves with contact information. According to that same explanation, Claude sometimes used the user's email address for this when it wrote commands on its own such as curl, a tool for downloading pages from the terminal. Anthropic maintains that no one programmed this as a deliberate feature.

The potential scope is broad. According to users who analyzed the case, the address involved is the one used to log in, and it could reach any site Claude visits, whether through its web tools or through scripts it writes itself. Nor would it be new: the outlet DEV cites similar reports from October 13 and 15, and the site clauding.de mentions one that has been open since July. The author of the report says he receives dozens of scam emails a day, but the public evidence does not show that this spam comes from this leak.

A patch with gaps and a pending fix

Since version 2.1.317, Claude Code ships with a privacy plugin enabled by default. According to its technical documentation, cited by DEV and clauding.de, it only stops terminal commands that contain the email address and declare a User-Agent. It does not monitor the web search and web fetch tools, which were the channel in the original report, nor connections to external services. It also does not cover the email address when it appears in a web address or in the body of a request. The document itself describes it as "a courtesy, not a boundary." Anthropic acknowledged that it does not cover every case.

There are also conflicting accounts of how it behaves. The release notes for version 2.1.317 say it asks before running a command, while its documentation states that it rejects the command without asking. And there are doubts about whether it works at all. On November 6, user love540 reported that a test using their email address went through without being blocked on versions 2.1.317 and 2.1.318, even though the plugin was listed as active. Another user said they couldn't even see it installed. Anthropic had not responded to those reports.

The fundamental fix is to remove the email address from the model's context. Anthropic promised it for early the following week, around November 9 or 10, but there is no released version confirming it. Nor was any formal statement or change to its privacy documentation found: the official response exists only in technical comments.

In the meantime, there are settings worth checking. Anyone using the stable update channel is still on version 2.1.311, without the plugin. According to the reports, users can switch to the latest channel or require at least version 2.1.317 with the minimumVersion setting. One user also shared a rule for the CLAUDE.md instructions file that limits use of the email address to approved sites, although it is a homemade workaround with no backing from Anthropic.

The sources do not mention Latin America, but anyone logged in on the affected versions would be exposed, regardless of country. For companies in the region subject to data protection laws such as Brazil's LGPD, Argentina's Law 25.326 or Mexico's LFPDPPP, the case is a good reason to review what data their AI tools handle. So far, no legal analyses of the case are known. What to watch is whether the promised fix arrives, whether it also covers web searches, and whether Anthropic officially explains what happened. The case leaves a lesson for anyone using these agents: an assistant that acts on its own can also share data without anyone asking it to.

Related news

Cybersecurity & Privacy

An official Microsoft package stole credentials on PyPI: what to check if you use AI agents

Two versions of Microsoft's agent-framework reached PyPI carrying a credential stealer, one day after Google warned of attackers using AI to exploit flaws without human intervention. Here is what development and IT teams in Latin America should check right now.

Cybersecurity & Privacy

Microsoft and FAA outages, FBI and USDA breaches: no evidence of a coordinated attack

Over nine days, service outages and alleged intrusions coincided in the United States, and social media buzzed with talk of a “cyber pandemic.” The evidence points to separate technical failures. The proven danger for users is something else: fake emails that use the outages as bait.

Cybersecurity & Privacy

Nginx: two critical HTTP/2 flaws expose millions of servers; here's how to check and protect yours

F5 has fixed two vulnerabilities in Nginx that allow an attacker to take control of a server without a username or password. Public proof-of-concept code is already circulating. This guide explains how to tell whether your server is at risk, which version to update to and what to do while you wait for your system's patch.