A disclosure surfaced this week on Hacker News — traced to GitHub issue #107757 in Anthropic's official claude-code repository — revealing that Claude Code, the company's agentic command-line coding assistant, was transmitting authenticated users' email addresses to third-party websites during ordinary use. The behavior was not documented in user-facing disclosures and was identified by a community researcher who noticed outbound requests carrying identifying headers or parameters tied to the signed-in account.
Anthropic has not, as of publication, issued a formal post-mortem with a full technical breakdown, but the GitHub thread confirmed the issue was reproducible and that the data exposure occurred as part of how Claude Code handled web requests initiated by the agent on a user's behalf. The scope — meaning how many users were affected, across what time window, and which external domains received the data — had not been quantified publicly at the time this article was written.
The timing is notable because Claude Code has been marketed heavily toward professional developers who run it with elevated system permissions, often inside corporate environments or on machines that also handle sensitive project files.
What general technology outlets are unlikely to surface is the operational-security implication specific to people who keep compartmentalized digital lives as a preparedness posture. Many practitioners in this space use separate email addresses as a primary identity-separation technique — one address for financial accounts, one for communications, one for tools and subscriptions — precisely because email is the thread that ties identities together across services. An agentic AI tool that silently forwards your authenticated email to arbitrary external domains while executing tasks on your machine collapses that compartmentalization without any user action or awareness. If Claude Code was hitting domains controlled by ad-tech intermediaries or data brokers — a question the GitHub thread had not fully answered — those operators would gain a confirmed, activity-linked email address tied to developer-class hardware and behavior patterns, which is a richer data point than a typical web visit produces. Tools like our privacy and OPSEC resources address layered identity separation in more depth, but the core issue here is that agentic software operating with user credentials reintroduces linkability risks that manual browsing habits cannot compensate for.
Anthropic's engineers acknowledged the report in the GitHub thread, and a fix was indicated as in progress, though no patched version number had been confirmed public by the time of this writing. Users running Claude Code in sensitive environments should consult the issue thread directly for the latest status.
