Anthropic's Claude Code could reveal users' email addresses during a normal internet search. One user complained that he was now receiving dozens of fraudulent emails every day.
Anthropic's Claude Code could reveal users' email addresses during a normal internet search. One user complained that he was now receiving dozens of fraudulent emails every day.
This is reported by Cyber Security News.
The story began on November 5th, when a GitHub user named taylor-edwards reported strange behavior from Claude Code. According to him, the AI assistant was sharing his email address with websites he visited while working without his permission. After that, he began receiving dozens of fraudulent emails a day.
“This shouldn’t be a setting that can be turned on or off. It needs to be removed completely. It doesn’t benefit anyone: it doesn’t help services that don’t want to receive requests from AI agents, it doesn’t preserve my anonymity on the Internet, but for scammers it’s a real goldmine,” he was indignant.
How did the email end up on third-party sites?
When a program requests a website, it can pass a special User-Agent header to it. Usually, it specifies which program is making the request. But Claude Code sometimes added his user's email address to this header.
Perhaps Anthropic itself gave Claude Code access to this information.
In a comment on Hacker News, a user named lydia_hallie, who identified herself as a Claude Code team member, explained that users had previously complained that the AI assistant would sometimes make up their names. For example, when it was necessary to indicate the author in comments to the code. To solve this problem, the developers began to transfer the email addresses of users who were logged in to their accounts to Claude Code.
But some sites ask automated programs to specify contact information. And when Claude Code himself created commands to access such resources, he sometimes added the user's email address to the User-Agent header that the site receives.
According to lydia_hallie, the developers are now planning to remove the account email address from the information the AI model receives. The corresponding update should be released early next week.
Anthropic does indeed have an employee named Lydia Hallie, but it has not yet been possible to confirm that the lydia_hallie account on Hacker News belongs to her.
It is not yet known how many people have been affected by the problem.
Earlier, dev.ua reported that hackers infected an update to the popular AI tool LiteLLM with a virus that stole passwords and other confidential data. During the investigation, the developers suspected that Claude Code, under certain conditions, could accidentally reveal secret GitHub access keys in its responses, which could have caused a hack.
