Security & Redaction
When I say “security”, I’m not talking about encryption or firewalls. I’m talking about:
- reducing cloud costs
- protecting the brand
- preventing users from abusing the system
- and avoiding accidental data leakage
First, we must ensure the LLM isn’t being used for off‑topic nonsense. A malicious user could pay for our product and then ask it to write Python scripts, compose love letters, or explain quantum physics. Guardrails exist, but as I showed in a previous post, they’re about as trustworthy as politicians.
Second, genealogists — bless them — often don’t care what data goes into the cloud. I’m guilty too. My Ancestry tree contains plenty of living people marked “private”, which is basically the digital equivalent of hiding something under the bed.
But from a brand perspective, I want to raise the bar. Even if users don’t care, we should.
Let’s illustrate.
Suppose you ask the LLM:
“Where was Jesus Christ born?”
Let’s not get into theological debates. For this post, he’s alive. Behave yourselves.
If I were descended from royalty, I might convince myself I descend from Adam and Eve, and therefore Jesus is in my tree. So when I ask that question, the LLM should respond:
get-data:Jesus Christ
My app then queries the GEDCOM and returns the data.
Now, Jesus is public domain. But what if the question was about your third cousin — the one you’ve never met but have definitely stalked on Facebook?
Type their name into the chat, and your horse hasn’t just bolted — it’s halfway to France on a ferry.
Last time I used that analogy, I forgot to include the image. So here it is.
[Image: courtesy of ChatGPT/OpenAI. Prompt: “Can you please create me an image, 500px x 500px, a horse on a ferry in the English Channel taking a selfie?”]
What a time to be alive.
Or it would be, if the world weren’t currently full of land grabs, bombs, and a Labour government.
