This news item from a few weeks ago revealed that when users created a “public link” of their Claude chat history, that chat history ended up in Google’s search queries.

Aren’t these urls supposed to be more or less undiscoverable by virtue of the long string of random characters that they contain?

How did Google discover those urls?

Did these people publish the urls somewhere publically themselves?

Or did Google “fish” them out of their gmail inbox or something like that?

  • SnoopSqueak@lemmy.today
    15·
    2 days ago

    Did these people publish the urls somewhere publically themselves?

    They created a public link. Google scraped it because it wasn’t told not to.

    From the article:

    A spokesman for Google made clear to the BBC that the company does not control “what pages are made public on the web,” saying instead that action comes from websites.

    “We give site owners clear controls to decide whether pages can be crawled or indexed, and we always respect those directives.”

    As the search indexing of the chat logs is no longer occurring, it is likely Anthropic used available tools to quickly block the chat log links from search results. Google’s process for a website owner to block a link is straightforward, but must be initiated by a website owner.

    Google was not the only one:

    Other search engines like Bing, Brave and Duck Duck Go, through which the Claude chat logs also appeared, were approached for comment.

    • khanas1f@lemmy.world
      4·
      1 day ago

      Anthropic blocked the scraping quickly when this came to light but does google still has the URLs that it has already scraped?

      I understand that new shared chats won’t show up but what about the old ones.

      • SnoopSqueak@lemmy.today
        1·
        1 day ago

        Yes, Anthropic presumably submitted a request to block search results that lead to Claude chats, as well as to remove the ones that were already there. Google’s removal tool can remove results that had been indexed, and the “noindex” meta tag prevents future indexing, if I’m understanding correctly.

        https://support.google.com/webmasters/answer/9689846?hl=en

        does google still has the URLs that it has already scraped?

        My guess would be yes. I doubt the removal tool actually deletes content, more likely it just deactivates it. It could theoretically be reactived or used for something else, possibly even inbreeding (AI chat logs to train AI on) or selling data to whoever pays for it. But that is just a guess, I could be wrong. I don’t have much optimism when it comes to corporations, they tend to be as evil as they can get away with. If they legally acquired the content from those links because Anthropic neglected to protect them, there may not be many limits to what Google can do with those chat logs.

    • KurtVonnegut@mander.xyzOP
      2·
      1 day ago

      Ah ok, looks like I didn’t read the article carefully enough.

      That still leaves two scenarios though:

      1. Claude fucked up, and gave Google a way to index public links of people. After this news broke, they asked Google (and other search engines) to not show those urls anymore.
      2. People fucked up and/or fell victim to Google’s the data hunger, which let Google to index their public links, after which Claude asked Google not to index such links.

      I guess, in the end, I’m mostly wondering if the long random strings of public links are generally “brute-force-discovered” by crawlers like Google, or whether “discovery” of such links requires some kind of breach.

      • SnoopSqueak@lemmy.today
        4·
        1 day ago

        Claude fucked up, and gave Google a way to index public links of people.

        Anthropic fucked up, Claude is not a person.

        After this news broke, they asked Google (and other search engines) to not show those urls anymore.

        Correct, that seems to be the case.

        People fucked up and/or fell victim to Google’s the data hunger, which let Google to index their public links

        Kind of. The user fuck up would be not realizing that Anthropic had no protections from search engines, and that by clicking “share” they were creating publicly accessible links.

        after which Claude asked Google not to index such links.

        Anthropic seems to have updated the domain those links are generated on so that crawlers do not index public chats, yes.

        I guess, in the end, I’m mostly wondering if the long random strings of public links are generally “brute-force-discovered” by crawlers like Google, or whether “discovery” of such links requires some kind of breach.

        I am also curious. I looked it up. https://developers.google.com/search/docs/fundamentals/how-search-works

        The first stage is finding out what pages exist on the web. There isn’t a central registry of all web pages, so Google must constantly look for new and updated pages and add them to its list of known pages. This process is called “URL discovery”. Some pages are known because Google has already visited them. Other pages are discovered when Google extracts a link from a known page to a new page: for example, a hub page, such as a category page, links to a new blog post.

        If I had to guess based on this, Google had probably indexed the domain Claude or its frontend run on and automatically scooped up the public chats as they became available.

        It is also possible that any Claude chat shared will still be accessible to web crawlers, if they just ignore the flags telling them not to scrape, crawl, access, index, etc.