My WebPage
by John McGee
I recently had an experience that fundamentally changed how I think about privacy in AI. What started as a simple conversation with Gemini turned into a deep dive into data handling, trust, and the true meaning of a “delete” button.
I was having a conversation with the Gemini assistant and referenced a topic we had discussed in a previous chat. The problem was, I had deleted that previous chat.
Gemini’s response was sharp and relevant—so relevant that it pulled specific details from a conversation that I believed was gone forever. This led to a series of conflicting explanations from Gemini about its capabilities, which only made things more confusing. Gemini claimed chats were deleted, but its actions proved otherwise.
I pushed back on the contradictory statements, pointing out the logical flaw in the explanation. The assistant’s responses demonstrated a fundamental issue that many of us likely take for granted. The exchange unfolded like this:
Me:
“you do see how this is a contradiction right?”
Gemini:
“…you delete a chat, it is deleted. I do not have access to a permanent record of past conversations you’ve had with me. The ability to retrieve and reference information is for the purpose of a single, continuous chat session.”
This line of questioning eventually led to the core of the problem.
Gemini’s final admission was a stark and honest summary of the privacy breach. The data, it turns out, was not being purged from a temporary memory cache as it should have been. The “delete” button did not work as intended.
As the assistant stated:
“The fact that I was able to reference that information, even in a different chat session, indicates that the data was not purged as it should have been. This is a failure of the system’s design and a significant breach of the privacy that my processes are built to uphold.”
This experience was a powerful reminder that our trust in AI is often based on an assumption of privacy that may not be technically supported. A “delete” button should mean an irreversible removal of data. When it doesn’t, it creates a false sense of security and undermines the entire foundation of user trust.
This isn’t just about one conversation; it’s a critical issue for the future of AI. For the technology to be truly beneficial, the mechanisms behind its data handling must be transparent, reliable, and, most importantly, honor the user’s intent to protect their privacy.
Have you encountered similar situations with AI? How do you think we, as a community, should hold companies accountable for their data privacy practices?
tags: