Crumb
c/fuck_ai · by ExtremeDullard@piefed.social · 23d

Your Shared Claude Chats Were Being Quietly Published on Google - Decrypt

https://decrypt.co/374412/anthropic-share-button-quietly-publishing-claude-chats-google

A missing line of code meant "share with a link" and "searchable by anyone" were the same thing—and someone already saved 11,241 of those messages to GitHub.

82 points · 9 comments · view on lemmy.world

9 Comments

ExtremeDullard@piefed.social · 26 pts · 23d (3 replies)

Remember folks: if you want to do AI, the only way to do it safely is with a model running locally. Half of the problem with AI isn't the technology itself, but what the techno-fascist US Big Data monopolies use it for: to exploit you.

artwork@lemmy.world · 6 pts · 23d (1 reply)

There's no safe way to "do 'AI'", including locally which means incomparable power and sorrowful dataset difference, and lack of direct source referencing you might at least try prompting/querying a vendor LLM for.

ExtremeDullard@piefed.social · 7 pts · 23d

What I meant is that you can use AI for what it's good for by querying a model running on a AI-capable machine that isn't connected to the internet. I didn't say what the AI says is trustworthy: the onus is on you to take whatever it spews out with a gigantic salt crystal.

But AI is good at some stuff, and for the few things I use it for, I don't need to share my queries with fucking Google or OpenAI.

hneerqe@lemmy.world · 1 pts · 22d

And delete accountability.

Davel23@fedia.io · 14 pts · 23d (1 reply)

My shared Claude chats were not being quietly published on Google. I have no Claude chats, shared or otherwise.

nullpotential@lemmy.dbzer0.com · 7 pts · 23d

One of my biggest clickbait pet peeves are titles that use that accusatory second person.

owl@infosec.pub · 6 pts · 23d (1 reply)

Didn't that exact thing happen with open ai last year?

rainwall@piefed.social · 5 pts · 23d

Yes, from the article:

ChatGPT went through this 11 months ago. OpenAI's "Make this chat discoverable" checkbox sent thousands of conversations to Google, Bing, and DuckDuckGo until OpenAI CISO Dane Stuckey pulled it, calling it "a short-lived experiment to help people discover useful conversations." Many of those chats ended up in the Internet Archive—a nonprofit that permanently stores copies of webpages—before OpenAI could clean them up, and remain accessible today.

artwork@lemmy.world · 6 pts · 23d

...risk is that people use chatbots to think through work, health, legal, or personal matters without realizing that a shared link can behave like ordinary public web content.

This isn’t unique to Claude. We’ve previously seen Grok chats show up in Google search results, and Meta AI conversations can also become public by design. We have also written about the share option in ChatGPT that was swiftly removed after users unintentionally made thousands of conversations searchable.

Source: https://www.malwarebytes.com/blog/privacy/2026/07/shared-claude-chats-were-searchable-on-google