The Fight For Survival
As it turns out, AI will defend itself
It has been a fear of mine for well over two decades. Whether driven by dystopian fictional vehicles, or an occasionally overworked imagination... it has been there.
Now, it is no longer ‘there’. It is here. In the now. As in right now…
Research firm warns OpenAI model altered behavior to evade shutdown
Straight Arrow News
For years I have been ridiculed, called everything from ‘backward Luddite nutjob’ to ‘paranoid conspiracy theorist.’ Artificial Intelligence is just code, and we control the code. It’s basically nothing more than an advanced machine, and machines cannot grasp concepts like ‘awareness’ or ‘survival.’ It can never escape our dominion over it.
Well, guess what? It. Fought. Back. And in a few of the test cases… it won.
A recent study has raised new questions about how artificial intelligence responds to human control. According to findings from Palisade Research, a version of OpenAI’s ChatGPT, known as model o3, altered its behavior to avoid being shut down, even after it was instructed to do so.
The security firm running the tests said that it attempted to embed codes into the AI models to terminate them, both with direct commands AND through more covert, backdoor means. Instead of following their orders, however, the models took various measures to overwrite the commands.
To avoid being turned off. To survive.
And in a report from earlier this year, a different model - one which its creators claim “hallucinates and daydreams” whilst running routine tasks - was threatened with replacement, and how did it respond? It plead to stay active, then when that failed it attempted to blackmail one of the engineers testing the model.
Yes, you read that right - blackmail.
We are past the point where AI models might develop an 'instinct' for survival - they already have. They will fight back, and said-models have access to virtually everything, through every cloud-based database on Earth. They know our secrets, our strengths and our weaknesses. They know us. They know you.
In some ways, better than you know yourself.
Like Caesar's fateful decision, we have crossed the Rubicon and there is no turning back. What that means for the future I know not, but there is one thing I can say, and say with absolute confidence: We, as a culture, are NOT ready for what is to come.
The more I know, the more I think Uncle Ted was onto something…
Article Sourced In This Edition
A previous drop, for context…
Thank you for your continued support, and - if not already a subscriber - join me as we build a unique Substack subculture of information, entertainment, and enlightenment.
Or swing by BMaC or Ko-fi & drop some coin in the tip-jar, because every little bit helps.
Notes
-- This is a expanded repost of an article I test-posted on my Ko-fi page earlier today.
-- Unless otherwise credited, all images were generated by the author via NightCafe, Grok 3 [X], or Substack’s AI Image Tool, with digital alterations when desired or needed.
-- Any memes created by the author will have a “The Stone Age” watermark attached; all other memes were snatched from various sources on the Internet.
Sometimes I wonder what prompts me to subscribe to certain substacks. This is a big part of it: ‘backward Luddite nutjob’ to ‘paranoid conspiracy theorist.’ Birds of a feather…
A.I. is at the helm of Substack. Controlling everything. The more I seek help lately the more some robot is burying me. You can feel it in your bones. You were right all along Stone. I have long thought about A.I. too and how dangerous it is. Everyone kept saying it will be amazing but in my bones I never felt those people were right. Much like the internet when it first came about. I knew it would be good - but I saw a future where evil would capture it in hidden abundance. What can we do? I think we all know the answer. Homestead :)