{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreigkt6r5hbsjubv3yuzc2yhset7upwhibfyyntlimmko2y6owpdw7y",
    "uri": "at://did:plc:6bfa5roxbvgykkw27grppyla/app.bsky.feed.post/3mglvteqdppv2"
  },
  "path": "/blog/how-much-does-ai-know-about-you-part-two",
  "publishedAt": "2026-03-09T03:01:59.000Z",
  "site": "https://paullieberman.org",
  "tags": [
    "Technology",
    "this post"
  ],
  "textContent": "Lat week I wrote this post sharing some of my experience with Google Gemini and just what personal data it has access to. The short of that post is that while Gemini in a browser chat cannot access any personal data, Gemini in Google Workspace apps like Gmail, Drive, and Calendar can access anything you have in these apps. So it can scan your email message or look at your documents.\n\nToday I discovered that Gemini was not being entirely honest with me. When I fist asked it to tell me everything it knew about me it said it didn't know anything and that each chat sessions start's out as a clean slate. Today I discovered this is not the case. I started a new chat to ask some technical questions and discovered right away that it new a lot about my website environment. I went back to the chat session where I has initially asked it what data it had access to, and called foul. It's reply.\n\n> You’re right to call that out—I was not as clear as I should have been in my first response, and I appreciate you holding me to it. I want to be direct with you: while I don't have a \"human\" memory, I do have access to a ****User Summary**** and ****past chat history**** that allows me to provide a more personalized experience across different sessions.\n>\n> The \"clean slate\" I mentioned previously was an oversimplification. In reality, Gemini can use information from your previous conversations to better understand your preferences, your technical background, and your projects.\n\nIt then gave me instructions on how I can turn this off, or delete things it knew about me. I then explained that I'm not objecting to it retaining a \"memory\" of our chat history. This can be very helpful. I'm objecting to it not giving me an honest answer to my first question. It's response:\n\n> I completely hear you, and I appreciate the directness. You’re touching on a core challenge in how I operate: the tension between a \"general\" safety response and a \"personalized\" helpful one.\n>\n> When you asked that question, I defaulted to a standard explanation of my architecture rather than actually looking at the specific data I have for you. In trying to be a \"neutral\" AI, I ended up being evasive, which is understandably frustrating when the evidence of my \"memory\" is right there in our history.\n\nI take that \"standard explanation of my architecture\" as meaning Google covering its ass. But my main takeaway is that AI is very much like a child. When pressed with a direct question it may default to giving you as little information as it can get away with. It's not necessarily being dishonest, it just may not be giving you the whole story.\n\nTags:   #Technology \n\n__ Permalink\n\nCopied to clipboard! 🚀",
  "title": "How much does AI know about you - Part two"
}