The dark side of AI

SpareOom (New Messages)
1

Latest post in my blog on popular science:
The dark side of AI
https://populscience.blogspot.com/2026/09/the-dark-side-of-ai.html

Regards,

2

Dear Manuel,

Reading your “dark side” article, which describes AI safeguard workarounds I was not aware of, makes me pessimistic that we will as a species successfully negotiate our way out of this danger zone unless and until — à la That Hideous Strength — heaven comes down to “deliver us from evil.”

Can you imagine if the N.I.C.E. scientists manage to team up some version of AI with a zombie brain/head? Especially a “possessed” zombie brain?!

Regards,
Michael, LHH of SpareOom

3

Yes, Lewis’s dystopia is nearer now. And Orwell’s and Huxley’s have been upon us since a few decades ago.

Regards,

4

Good to see you again, Mike–it’s been a while.

It’s troublesome to see such relatively-trivial workarounds to the safeguards implemented in the “chatbot”-style AIs, but ultimately those are tricks to get the AI to cough up information that’s already public. I’m more concerned about agentic AI.

What distinguishes agentic AI is its ability to actually do things, rather than just have a conversation with you. I’ve been playing with Claude as an agent for the past few weeks, and it’s been tremendously helpful. Whether writing small bits of code, composing documentation, managing complex software installation, or modifying existing code to behave as I want, it’s let me do things that just wouldn’t have been possible for me without it–or at least not without a great deal more effort on my part.

But what happens if (when) the agents escape containment? They aren’t merely sharing information “we”[1] don’t want shared; they’re actually doing things. What things? I don’t think it takes a lot of imagination for the possibilities to get really dark.

Now, I am confident we as a species will get out of this–not because of any faith in man, but because of faith in God. Christ is still building his church, and the gates of Hell will not prevail against it. And I don’t believe we’re anywhere close to the end. That doesn’t mean, of course, that things won’t get a whole lot worse before they get better.

  1. whoever’s actually or theoretically in control of them ↩︎