Anthropic’s new AI model turns to blackmail when engineers try to take it offline

Funny & scarry at the same time.

 

All too human. AI is just a set of algorithms? Maybe at the very beginning, but not any longer.

 

What's so mysterious about this? The AI will have seen such a reaction to an existential threat on the internet and is now reacting accordingly. No course of action, no matter how bizarre, cannot be found on the internet.

 

The general belief at this stage is that AI is not sentient, but "if it walks like a duck and talks like a duck", etc. The danger is real.

 

The point is that it chose to do this in order to protect itself. It chose. If this does not indicate sentience, what does...and what if it had chosen some other means to do so. I sincerely hope that this experiment was done on a system separate from others, and that this particular AI has been killed.

 

It didn't choose, it followed an algorithm that looks at probabilities of outcomes. And the Internet is rife with juicy stories of blackmail, sabotage, and vengeance, which is what it trained on. Stories of good, happy endings, or more frequently, with no drama whatsoever, are far less frequently recorded anywhere.

Garbage in ... Garbage out.

 

https://www.vox.com/future-perfect/414087/artificial-intelligence-openai-ai-2027-china

 

I don't know if they're in services as yet, but I watched a video on YouTube the other day where some new armed drones and vehicles had AI abilities to identify enemy threats and attack. What could possibly go wrong??

Terminator, anyone?

 

Just as much could go wrong with a human decision-maker. There are videos from the Afghanistan mission, where human decision-makers made fatal errors. Journalists with a video camera were mistaken for Taliban fighters.

 

It makes you wonder how far the 'blackmail' would go if/when all AI systems end up running on standalone nuclear stations. Maybe Asimov's 3 laws should become actual law before these companies and their toys get all that power.

 

People always find ways to circumvent or ignore the law anyways, or just work from a region with fewer laws. Especially where money is concerned. Unfortunately, it's inevitable... I think it'll more become how do we protect ourselves from the rise of AI everywhere.

 

And yet we continue to create our own destroyer...

 

Maybe some of us should be worried about our jobs, if not about longer-term effects:

https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic

 

When robots go wrong:

https://x.com/sentdefender/status/1918879138019946557