Google Exposes User Private Keys and Company Secrets in Claude Link Configuration Failure
Anthropic’s Claude AI recently faced a serious privacy breach: numerous user-generated session sharing links were publicly indexed by major search engines like Google and Bing, allowing anyone to view conversation history directly via search. This was not a backend database hack, but a technical configuration error by Anthropic—failing to block search engine crawlers from accessing these links.

The core issue stems from a disconnect between users’ perception of “sharing” and internet reality. Many users assume sharing a link with colleagues or friends is safe, unaware that once the link appears on public websites, forums, or social media, search engine crawlers can follow it and index the content.
Tests revealed that indexed sessions contained sensitive data—wallet private keys, illegal consultation records, internal company tasks, and personally identifiable information—all exposed in search results.
Technical misconfiguration leads to severe consequences: conflicting settings
The root cause was a contradictory technical setup. The /share/ path in Claude’s robots.txt disallowed crawling, yet the page response header also included X-Robots-Tag: none (equivalent to noindex, nofollow). While this seems like a double safeguard, it backfired. Google’s official documentation states that if a page is blocked by robots.txt, crawlers cannot read the noindex tag. As a result, crawlers fail to access the page content but still discover the URL via external links and index it. Users can then click the link to view the full session.
Anthropic has since contacted major search engines to remove the links and updated its strategy to fully block crawling. However, some cached or residual links remain. This is not an isolated incident—many ChatGPT sharing links in 2025 were also indexed by Google, and earlier, Baidu Netdisk’s public sharing feature caused large-scale exposure of user files due to similar issues. After repeated failures, a critical question arises: AI companies emphasize security and privacy, yet fail to manage basic technical configurations. Why should users trust them?
Related article
How Unitree is Shaping the Future of Humanoid Robotics
Unitree’s New Creature of Embodied AI with ultra-wide 4D LiDAR technology for advanced real-world navigation. Credit: UnitreeWang Xingxing, CEO of Unitree, targets world model breakthroughs to help humanoid machines execute 80% of tasks when placed i
Why Abnormal AI Partnered With OpenAI on Daybreak
Michael Aiello, Head of Product for Cyber at OpenAI | Credit: Michael Aiello/LinkedInAbnormal AI joins OpenAI’s Daybreak programme, as Mike Aiello says the partnership will accelerate the adoption of practical defences for enterprisesLeading frontier
LFM2.5 DSpark Draft Model Released: Inference Speed Boosted 3.18x
Liquid AI and Hugging Face have officially released the DSpark draft model checkpoints for the LFM2.5 series, covering LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B. This innovation introduces a novel speculative decoding path that significant
Related Special Topic Recommendations
Comments (0)
0/500
Anthropic’s Claude AI recently faced a serious privacy breach: numerous user-generated session sharing links were publicly indexed by major search engines like Google and Bing, allowing anyone to view conversation history directly via search. This was not a backend database hack, but a technical configuration error by Anthropic—failing to block search engine crawlers from accessing these links.

The core issue stems from a disconnect between users’ perception of “sharing” and internet reality. Many users assume sharing a link with colleagues or friends is safe, unaware that once the link appears on public websites, forums, or social media, search engine crawlers can follow it and index the content.
Tests revealed that indexed sessions contained sensitive data—wallet private keys, illegal consultation records, internal company tasks, and personally identifiable information—all exposed in search results.
Technical misconfiguration leads to severe consequences: conflicting settings
The root cause was a contradictory technical setup. The /share/ path in Claude’s robots.txt disallowed crawling, yet the page response header also included X-Robots-Tag: none (equivalent to noindex, nofollow). While this seems like a double safeguard, it backfired. Google’s official documentation states that if a page is blocked by robots.txt, crawlers cannot read the noindex tag. As a result, crawlers fail to access the page content but still discover the URL via external links and index it. Users can then click the link to view the full session.
Anthropic has since contacted major search engines to remove the links and updated its strategy to fully block crawling. However, some cached or residual links remain. This is not an isolated incident—many ChatGPT sharing links in 2025 were also indexed by Google, and earlier, Baidu Netdisk’s public sharing feature caused large-scale exposure of user files due to similar issues. After repeated failures, a critical question arises: AI companies emphasize security and privacy, yet fail to manage basic technical configurations. Why should users trust them?
How Unitree is Shaping the Future of Humanoid Robotics
Unitree’s New Creature of Embodied AI with ultra-wide 4D LiDAR technology for advanced real-world navigation. Credit: UnitreeWang Xingxing, CEO of Unitree, targets world model breakthroughs to help humanoid machines execute 80% of tasks when placed i
Why Abnormal AI Partnered With OpenAI on Daybreak
Michael Aiello, Head of Product for Cyber at OpenAI | Credit: Michael Aiello/LinkedInAbnormal AI joins OpenAI’s Daybreak programme, as Mike Aiello says the partnership will accelerate the adoption of practical defences for enterprisesLeading frontier
LFM2.5 DSpark Draft Model Released: Inference Speed Boosted 3.18x
Liquid AI and Hugging Face have officially released the DSpark draft model checkpoints for the LFM2.5 series, covering LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B. This innovation introduces a novel speculative decoding path that significant





Home






