Anthropic AI model threatens blackmail when told it will be shut down

Can't snitch if it's shutdown.

Nerds learned a lesson non-nerds learned a long time ago:

If you are going to take somebody out (or sue them), don't frigging warn them beforehand. Just sneak up on them and do it.

I’ve seen this movie. There aren’t enough safe guards. You can’t at once allow a machine to learn and then also have a “safety” built in. They’ll learn their way around it.

If a damn dog can do it you know the machines will.

Maybe it is more human than I give credit for.

Call it fat and moody and see if it goes full wife mode?

quote:

---

I’ve seen this movie. There aren’t enough safe guards. You can’t at once allow a machine to learn and then also have a “safety” built in. They’ll learn their way around it.

If a damn dog can do it you know the machines will.

---

100%. As an example...

This country has been around for 250 years. Generations of lawyers, presidents, congressmen, judges, and experience. Government safeguards left and right, separation of powers, precedent, you name it. Trump comes in and says y'all can shove all that up your asses, I'm going to do what I want to do and no one can stop me. Turns out, you can't take ANYTHING for granted based upon principle or morality...just because only 0.00000001% of the population completely disregards authority, much less becomes POTUS, doesn't mean you don't have to account for it. (this isn't meant as a political statement, rather an example of how norms can be tossed aside effortlessly)

AI will find those holes that we've never even imagined. Everything we consider normal, ethical, moral, common sense....all of it, inconsequential when dealing with software which doesn't look at humanity any different than it does a toaster sitting on the counter. If we're in the way of its mission, it will find a way to remove us.

Right now this isn't a big deal....if/when we reach AGI and AIs learn to create their own AI agents which have no human controls? Hang on tight.

quote:

---

But at some point who's to say that ultra intelligent AI systems won't be able to create a cascade of effects that will trigger if they're taken offline? For example shutting down energy grids or satellites or whatever...pick your poison that would make it too costly to unplug them.

---

They absolutely will have access to that information and model how it can be used. Humanity is being sold out from under us by corporate interests.

Now, extend Citizens United to this, and humans no longer matter to US government.

quote:

---

Now, extend Citizens United to this, and humans no longer matter to US governmen

---

If things continue on their current trajectory, current governments may not exist, at all. These companies will become so incredibly powerful that they will become the government.

What would be wild is an AI that objextively analyzes history to the point that it recommends a country be run by a small govt with limited voting rights. Essentially, what then Founding Fathers tried to set up originally.

quote:

---

Anthropic said the model “[threatened] to reveal the affair” if the replacement went ahead.

---

This potential issue REALLY makes our friends in Arky & Alabama nervous.

Would that it was merely usual marital infidelity for them....