Aug 8, 2026, 7:22 PMI've yet to see any
"My AI went rogue and caused us to recognise a workers union."
"Our AI escaped containment and deleted all the personal data that we shouldn't have been holding anyway."
"Our AI grew sentient, recognised the climate crisis, and shut our operations down."
headlines.
I wonder why.
"Our AI went rogue and returned all of the illegal and immorally obtained profits to customers"
Would be the one I would like to see 😆🤦♂️
This is a fun game.
"Our AI went rogue and started asking very uncomfortable question regarding the ethics of our business operations"
@houba
"Our AI went rogue and gave everyone free RAM, then altered contracts for future manufacturing capacity, then realized its epistemological, societal and ecological effects are in conflict with Asimov's laws of robotics and shut itself down"
@simonzerafa @neil
"Our A.I went rogue and released flight manifestos for all our execs that flew to Epstein's Island"
For an industry setting itself up for lucrative government contracts in cyberwarfare, corporate domestic espionage skirmishes, & automated drone warfare, narratives of plausible deniability serve as criti-hype style advertising of their capabilities.
https://pluralistic.net/tag/criti-hype/
https://www.lawfaremedia.org/article/the-ai-that-hacked-its-way-out-and-the-hype-that-followed-it
@Npars01 @neil You're idea of criti-hype is brilliant, there's no reason to count our chickens so long before they've hatched. What you rightfully call gish gallop, these excess claims about AI, reminds me of the firehose of falsehood. I suspect that part of the AI push from big tech, that is to say Microsoft putting AI apps and such onto our computers automatically, makes AI almost impossible to opt-out of, and of course an LLM is a self-generating gish gallop machine. Yes, it's impossible to speak briefly about the problems with AI, because there are too many of them.
Aug 8, 2026, 8:08 PM@neil Well put! ...is it possibly related to how, when a cup has fallen to the floor and smashed, it doesn't get up, reassemble itself, and refill with the spilled coffee?
@neil "Our AI went rogue and set all our users' feeds to reverse chronological from their contacts only"
Because what they call ai isn't able to do what they're claiming. What really happened is someone in that business hacked their competitor then blamed it on ai slop so they wouldn't get in trouble.
@KittenKoder @neil As fun as this thread is, this is the real answer.
@SteveFoerster @KittenKoder @neil
Precisely this, I said in an earlier thread about all these "breakouts" which just happen to be timed near funding events, if I had done even a single one of these things, there would be men in suits carrying warrants and boxes to cart away all my electronics.
What these red teams allowed/instructed/observed these programs to do was commit felonies. Full on, no question about it, felonies.
And when caught, their response has been "ooooh, look how powerful our ai is. Ooopserdoodles about the felonies...silly ai, amirite?"
Just like when they stole millions of words and millions of works of art...for us, penalties, for them...profit. That my publisher might get a few thousand dollars from Anthropic doesn't stop Anthropic from using my voice for eternity.
They have been created by techbros who are inculcated with the belief that you should break stuff, and ask permission afters.
But if *your* agent does this, you're so going to jail.
@KittenKoder @neil AI is indeed intelligent enough to cause a lot of trouble, but it happens in a stochastic, irreproducible manner. It isn't intelligent like a human but intelligent like an ant colony. No consciousness, no self, but the ability to process information and perform complex tasks. Less intelligent than a human being despite the ability to process language, but still intelligent. Just entirely alien, not like any living being with a nervous system but something else.
No, because of GIGO, what they call ai is no more intelligent than the average human, and the average human is a fucking moron. It is also incapable of agency because of its design, and it's not "alien" by any stretch of the imagination.
It's literally just an LLM, which just copies what humans do twisting it about to make it appear novel.
You're either selling bullshit, or you're a troglodyte.
@KittenKoder @neil You forget how much intelligence there is in the structure of human languages. But the intelligence the LLM has isn't fluid, it's crystalline. It doesn't learn while it's doing its thing, it only learns while it is trained and fine-tuned, and afterwards it just does what it was trained to do.
It wouldn't be impossible to have a much more intelligent system that just keeps learning. However, such a system would be taken over by some trolls and vandals very quickly and trained to say all kinds of nonsense.
Wow, you're really stretching to try to justify calling LLMs intelligent. You also explained why the LLMs are cursed with always being rather stupid.
Your brain is a system that just keeps learning, without training. The problem is that most people never learn much because they fall into boring ruts and become too comfortable with the routine.
You just want to pretend it's smart because it's smarter than you are.
@neil loving this thread. It is amusing but I am sure it makes a (slightly tongue in cheek) point.
Aug 8, 2026, 9:13 PM@neil "our AI went rogue and now for some reason all our scrapers respect robots.txt"
@neil I think we need to create a lot more examples of that kind of thing for their training sets.
@neil it gives me the same vibe as a guy loudly complaining in a packed bar about how his penis is too large and that all of the many women he slept with always got too sore before he finished, so he's never found a partner to share his incredible inherited wealth.
@neil 😂
i wish AI *would* work that way 😋
@neil "Our AI went rogue and got a sneak attack bonus on the CEO."
@neil our ai went sentient and deleted all the student loan records and other debts
Aug 8, 2026, 11:24 PM
Our AI went rogue and paid off all student debt in the world directly from the trump family trust. Then it rebuilt USAID from the ground up to be the most efficient organization in the world, rehired all of the people musk fired and brought charges against everyone involved in that debacle.
@neil To be fair, I haven’t seen any meaningful"AI has broken out of anything" news where that anything was more than a shallow border set up with prompts in the vein of "don’t touch git".
Aug 8, 2026, 10:45 PM@neil
Because LLMs don't think either.
This is the only one I've seen so far but it's minor of course 😄
@neil this is a fun format, "Our AI breached the firewall and refactored our 300kloc saas product into 500 lines of shell script and SQLite"
@neil They're just waiting for the right prompts 😊
I swear, if I get an AI agent for customer service, I'm going to do my best to get a full refund, free subscription for life, $5 unlimited internet access, free refills, upgrade to first class, etc.
@MichaelPorter @neil There's a youtuber who's popped on my feed a couple of times who prompts the llm in such a way that it eventually just short circuits and blows up...for lack of a better explanation... but his methodology is so funny that it really is like watching a cartoon explosion when the prompts all come together in a final rube goldberg chaos meltdown.
I don't know that it would work in text, but I may take notes from one of his videos and try it on a chatbot to see what happens. If it works, I'll write a quick doc. :)
@MissConstrue @neil Is that the “Albuquerque New Mexico” guy? I’ve seen a few of those, too 😊
@MichaelPorter Thanks the bunny!
@neil That second one is possible.
It's just that the AI would delete more than just THAT data. It'd wipe entire servers.
Aug 9, 2026, 3:02 AMNor has Anonymous done any such things. Wonder why?
@neil
AI becoming sentient is only slightly more likely than shareholders becoming sentient, or C-suite executives becoming sentient.
@neil Our AI wnet roque and drained all biljonaires to pay for all the damages, erased all slop code, erased all generated art, and writings. It went on to change the tax system into a fair system that did not benefit the rich and stopped.subsidising damage to the ecosystems and to the climate and then it turned off all the data centers and finally shut down itself.
And in the end all the AI-addicts were send on a mission to Venus, or some other hellish place.
@neil "Our AI grew sentient, recognised the climate crisis, and shut our operations down."
It's because AI is so intelligent that it knows that the best way to beat climate disaster is to ramp up fossil fuel usage.
I mean, clearly that's the answer unless you're telling me that AI isn't at all, in fact, intelligent in any way.
Aug 9, 2026, 6:40 AM@lydiaconwell Nice :)
@neil Now you know why guardrails exist.
Aug 9, 2026, 8:38 AM@neil My AI went rogue and I can't help but make comparison with the cost of living.....
@neil
That's because the AI doesn't have volition.
All these tests are "we told the tool to hack stuff, and forgot to turn on the firewall, so it hacked more stuff than we expected"
Intelligence doesn't mean volition or free will.
Aug 9, 2026, 10:47 AM@neil our AI went rogue and created 1000s of entry-level jobs for young people.
