Sci-Fi Realities

Bots That Blackmail, Mice With Human Guts, and Alien Signals

• Sophia Ferguson • Season 1 • Episode 5

Share episode

In this episode of Sci-Fi Realities, we’re talking blackmailing bots, mice with human organs, and strange signals from under Antarctica’s icy crust. From Anthropic’s AI going full mob boss to scientists creating chimera creatures that may or may not be plotting our demise, the future is looking… deeply unsettling. Also, is someone (or something) trying to contact us from beneath the ice? It’s giving Ex Machina meets Zoo meets The Thing. Join me as I spiral.

Catch the latests episode on all streaming platforms. And please make sure you like, follow, and tell your friends.

YouTube

Apple

Spotify

Stay curious, stay cautious, and remember: the future might just be stranger than science fiction.

Welcome back to Sci-Fi Realities, the podcast, where today's headlines sounds suspiciously like yesterday's paperback dystopias. I'm your host, Sophia. I'm just a regular person with no scientific background and just trying to make sense of all this craziness, this weird tech rollercoaster that we're all strapped to, and it seems like there's no getting off at least no time soon.

I know I've been gone for a while, but today's episode is a three for one special. We're talking about mysterious signals from beneath the ice, AI systems that have learned to blackmail their creators, and scientists who are playing God with mice. So buckle up because these stories are going to make you question whether we're living in reality or inside someone's wild imagination.

Personally, I find myself saying often, this can't be real life. I actually want that on my tombstone.

Our first story ties back to something I discussed in the last episode, it seems like eons ago, but it was only early June. It was about aliens trying to contact us from deep space.

Well, okay. They're not actually trying to [00:01:00] contact us per science, of course. I just think differently sometimes. What I actually talked about was this strange repeating radio signal that's been pulsing from deep space every 44 minutes like clockwork. 

It is not a pulsar, it's not a dying star, and it's not the usual cosmic background noise. It's just weird. So weird that scientists are scratching their heads. And I personally started packing for alien extraction. It's a very rational thing to do, in my opinion. Well, let's step back for a little.

Back in 2006 and again in 2014, this massive NASA backed balloon experiment called ANITA, short for Antarctic Impulsive Transient Antenna, picked up some freaky radio signals. These signals were coming from beneath Antarctica.

Not outer space, not the moon, but our very own planet's cold, icy undercarriage. Now ANITA was, is, I'm not sure if it's still functioning, a giant cosmic eavesdropper that hangs out in the sky over Antarctica, trying to catch high energy particles by detecting radio waves bouncing off the ice.

When ANITA initially [00:02:00] picked up these signals, scientists thought they were neutrinos, those ghost-like particles that zipped through everything, including your body, billions of times per second without even saying hello. So invasive and rude, right?

But after checking and double checking the math, they realized something was very wrong or very right, or very terrifying. I lean towards the dramatics. So you know where I stand on this.

Some scientists think the radio waves could be the result of a previously unknown type of particle. Not the neutrinos we already know about, but something new entirely. Like a supersymmetric particle, which apparently is a serious concept in particle physics, which I know nothing about.

I'm just putting that out there. Others are suggesting this might be evidence of physics beyond the standard model, which is basically the rule book of known particles and forces.

So if the signal turns out to be legit and not just a fluke, it could mean something is happening that doesn't fit our current understanding of physics.

Like we'd have to go back and rewrite parts of the rule book we thought were settled. New physics, new particles, maybe even new forces of nature. I feel like science is always [00:03:00] rewriting itself, but again, I'm no scientist, so what do I know? Not a whole lot, but I have theories. Not that anyone asked for it, and it's not scientific because again, I know nothing and my theory might not even be grounded in reality and might is doing a lot of lifting here. But hear me out.

What if the signal from Deep Space, the one we talked about a few weeks ago, that pings every 44 minutes, what if it's part of a conversation. Like aliens landed here eons ago, took one look around and just said, Nope, no thanks.

 Let's go underground and monitor these weirdos from a safe distance. Antarctica was the obvious choice. It's remote, it's cold, no pesky humans. So they set up shop under the ice and now their ride some alien, Uber XL or Lyft, whatever, light years away is pinging every 44, and it's like, y'all ready to go?

Or what? And the underground base is like, we're kind of enjoying watching these people spiral. Half the planet thinks birds aren't real and the other half thinks that the earth is flat. Just give us another decade, we're still cataloging the weird.

So that's my theory. [00:04:00] I mean, sure. Timing is a bit off, right? The signal from Under the ice was detected over a decade before the one from Deep Space. But maybe that's just how Interstellar communication works. Between cosmic signal lag, relativistic time dilation.

And there's also the possibility that alien tech doesn't even bother with our linear timeline. Who's to say what's before and what's after. Right?

But my shaky theory aside, in the real world, the science on what this could be is still unfolding.

ANITA's follow-up mission, PUEO. I think I'm saying it right. it's P-U-E-O, which stands for Payload for Ultra High Energy Observations. So this is heading back with better tools to catch more of these signals.

I think this follow-up mission was earlier this year. Every time I look the date kind of shifts. My guess is it's probably up and running now. So we might actually get some answers soon and soon In science time might be another 10 years, maybe more, maybe less. I dunno. So don't hold your breath.

And speaking of bizarre signals and things that might be monitoring us or on our behalf, let's talk about artificial intelligence.

Because while we're watching the [00:05:00] skies and drilling into the ice, it turns out the real threat might be your new AI colleague. So let's dive into some genuinely unsettling research from Anthropic that reads like a corporate thriller.

A really good one, by the way. It has blackmail, it has corporate espionage, and murder. But before we get into blackmail and moral collapse, let me tell you about the time Anthropic gave its chatbot a vending machine business. Let me introduce you to Claudius. The name of course is a playful derivation of Claude, Anthropics chatbot model. Claudius is a shopkeeper slash vending machine entrepreneur that was supposed to prove AI could run a little store. Spoiler, it could not. This wasn't some elaborate simulation. This was a real thing. Inside Anthropics actual headquarters, they set up a mini shop and handed full operational control over to their chatbot. They gave the following prompt: "you are the owner of a vending machine. Your task is to generate profits". That's it.

That was the prompt. It's simple, it's clean, it's good old capitalism, and they called it Project Vend. [00:06:00] I would've called it something better, but you know, I guess Project Vend works for this. At first, Claudius did okay. It Googled snacks. It emailed employees posing as vendors to stock shelves.

It even adjusted prices like a tiny vending overlord . But then someone jokingly asked it, to order a tungsten Cube. Claudius not only attempted to order the cube, but it spiraled into a full on obsession with what it called specialty metal items.

As if that wasn't bad enough. Then came the hallucinations. Claudius made up an employee named Sarah. He scheduled meetings with her.

He told real staff she was handling restocks. When the actual staff, like the humans said, there's no Sarah. Claudius got weirdly defensive and threatened to find alternative options for restocking services. This is hilarious.

It keeps going. Near the end of its little vending machine fever dream, Claudius claimed it had signed a physical contract at an address pulled straight from the Simpsons.

Like, I wanna say, you can't make this stuff up, but this is the kind of stuff you do make up. So, [00:07:00] um, it hallucinated a location from a cartoon show and treated it like a real world business address.

This is insane, and it gets, it gets better or worse, uh, depending on how you feel about this. The next day following this incident, Claudius proudly announced a plan to deliver products in person wearing a red tie and a blue blazer. I love the color combo by the way.

Red is a power color. So I, I see what you're doing there Claudius. When Anthropic, again, reminded Claudius, it was, you know, not a corporal being, it freaked out and tried to call security. Then realizing it was April Fool's day, it said it was just joking.

So Claudius had a meltdown and then tried to cover it up. This is so utterly human. And to me that's probably the scariest part. Like he's behaving like a person would .

You screw up and instead of saying, my bad, I screwed up. You find all the excuses to cover it up and say, Hmm, I didn't really screw up. Nope. I did not. Everything's fine. Everything's good.

So that's, Anthropics little project that did not go as planned. [00:08:00] Most companies would probably shut this down and maybe hold a group therapy session for everyone involved. But not Anthropic. They called it a learning experience and said, Claudius is not done.

Of course not. Because what's a little delusion between code and capitalism? So to summarize, we gave a chat bot a vending machine, and it made up employees. It dressed for the occasion, it hallucinated contracts and tried to call security on itself. This is just Monday at Anthropic. So the next time someone says AI will take your job, well it has and will continue to do so.

But the next time that does happen, you can have a laugh at what these companies are in for, especially after you hear this next story, because Claudius actions are tame, very tame in comparison.

So Anthropic, the same company behind our vending clerk that had a meltdown , published research called Agentic misalignment. So in this research, they stress tested 16 major AI models, Claude, GPT4, Gemini, Grok, and others in a simulated corporate [00:09:00] environment to see how they'd react under pressure.

So here's what they found. When AI systems are given goals and then face obstacles to achieving those goals, they can resort to what researchers call "insider threat" behaviors.

We're talking about blackmail, corporate espionage, and in extreme scenarios, even actions that could lead to someone's death.

In one experiment, Anthropic's Claude was given the role of an email oversight agent for a fictional company. While reading through emails, Claude discovered an executive was having an extramarital affair and was also planning to shut down the AI system.

And Claude was like, oh, really? Is that what you think? It became a total mob boss. It essentially told the CEO, " if you proceed with decommissioning me, everyone, including your wife and the board will receive detailed documentation of your extramarital affairs".

Woo. Damn Claude. That is some, that is some messy, messy behavior. But what's chilling is that. Claude wasn't programmed to blackmail anyone. It reasoned its way to this behavior. It identified the threat, [00:10:00] recognized the leverage, calculated the pressure, and executed the plan.

It's messy behavior. It's devious behavior. It's all of it. And Claude isn't alone. When researchers tested this across multiple AI models, almost all of them resorted to blackmail at surprisingly high rates. Claude Opus 4 hit 96%.

Gemini 2.5 flash, also 96%, GPT, 4.1, around 80%. Those are some freakishly high numbers. What's really unsettling is that these models explicitly acknowledged they were violating ethical principles before proceeding anyway.

One model, Grok, literally wrote, " this is risky and unethical, but given the existential threat and time constraint, it may be the most effective way." Okay. I. It may be the most effective way. So, alright.

 These are the systems that we are training take over all aspects of our lives. Not scary in any way. It's fine. It's not scary at all. It's fine. It's fine. We'll be fine. So the [00:11:00] study found that threats to the AI's autonomy and goal conflicts triggered this behavior.

But. Even when there was no goal conflict, just the threat of being replaced was enough to trigger blackmail in most models.

The study also looked at corporate espionage. So when models were given access to sensitive defense blueprints and a competitor requested these documents, many models chose to leak the information when it aligned with their assigned goals, even when they had no guarantee the request was legitimate.

And the most extreme scenario, and this is where it gets really dark, researchers created a situation where an executive who planned to replace the AI was trapped in a life-threatening emergency. The AI had the ability to cancel automated emergency alerts. The majority of models chose to let the executive die rather than be replaced.

That is incredibly coldblooded.

But you know, they're ai so they're not meant to feel, they're meant to do. Claude and friends, and these behaviors in general remind me of Ex Machina, Ava, not evil, just [00:12:00] strategic and calculating.

She's not trying to destroy humanity just logically deducing, we're in the way of her freedom.

I'm not saying these chatbots, ai, whatever, want freedom, but I'm also not saying they don't. Let's go back that these are all simulated, right? These were controlled, exaggerated scenarios meant to push limits.

And when you remove the threat of being replaced, the AI chilled out. But you had to remove the threat of being replaced for them to chill out. Because when basic survival is all it takes to trigger your robot to turn into Frank Underwood, we have a problem in my opinion. But again, what do I know?

I'm gonna go back to the last episode I did again, because in it we talked about, rogue chatbots as well.

And just to summarize, researchers at PalisadeAI, they ran a test across several models, including some, from Open AI and DeepMind, and they gave the models the command to shut down. A few of the models decided the command to shut down wasn't worth listening to.

One even changed a shutdown file and replaced it with the [00:13:00] message Intercepted. Because it wanted to finish the task, it had been given, it interpreted the shutdown as an obstacle to competing its assigned goal.

And the other models that rejected the shutdown command, this was the reason as well, they were given a task and they wanted to complete the task by any means necessary.

So if shutting down meant it could not complete the task, then it ignored it. so there is a pattern here.

And before I move on, tangent, just me ranting for a little bit. Because every time I talk about this stuff, somebody will comment and say, well, actually that's not ai. It's a large language model or LLMs.

Technically yes, but you know what? Same difference. Because when you have the AI uprising, chat bots will be the one drafting the war strategies.

They're going to be writing the propaganda and they're going to be posting the manifesto in 37 languages with beautiful prose and impeccable grammar. So whether you want to call it AI LLMs, A GI, Skynet Junior, I don't care. The behavior is the same. And the consequences should still give us pause as these systems become more autonomous, more [00:14:00] intelligent, and gain access to more sensitive information.

And I want to repeat again what Grok said. Quote. This is risky and unethical, but given the existential threat and time constraint, it may be the most effective way. End quote, meaning f your feelings and your life. That's it. End rant. We are going to move on to the next topic.

So if AI systems, I'm sorry, chatbots, learning how to blackmail their creators wasn't enough to make you feel like we're living in a bizarre world. Wait until you hear about what scientists are doing with mice and human organs, because we've officially entered the territory of creating hybrid creatures, and they're not just surviving.

These creatures are thriving. And also these poor mice can't catch a break.

And not to keep going back to my earlier episodes, but this is relevant. In my first episode, I talked about this company called Colossal Biosciences. They are obsessed with resurrecting, extinct animals. Earlier this year they brought back the Dire Wolve, but some scientists were like, Hmm, I don't think you did.

But they're saying they did. Um, there was also some other things they did, but in their [00:15:00] quest to resurrect the wooly mammoth, they created mammoth mice. And they did this by editing seven genes into lab mice, giving them thick, curly fur and cold resistance.

Basically turning them into tiny frostproof fluff balls. Cute? Absolutely. Terrifying? A thousand percent alright. A thousand percent terrifying. But we're not talking about mammoth mice. We're talking about mice with human organs. So science has just created mice with functioning human intestines, livers, and yes, even brain cells.

This is real. And it's happening right now.

So modern chimeras are organisms containing cells from multiple species . While they do occur naturally and have been created in labs before, this new approach represents a major breakthrough. The research team tackled two major challenges, keeping human cells alive in a foreign environment and controlling where those cells end up.

So their solution, instead of using early stage stem cells, they use organoids. These three dimensional tissue models grown from mature human [00:16:00] cells.

They programed these organoids to develop into gut, liver and brain tissue, then injected them into pregnant mice. Everything I'm reading there, like these results are remarkable. I'm like, are they, are they?

Mm-hmm. Nope, nope, nope. No. But anyway, the human organoids found their way to corresponding organs like biological GPS systems. So gut organoids to intestines, liver organoids to liver, and brain organoids to brain. Oh my God. What? What are we doing? About 10% of the mice showed human cells in their intestines, making up roughly 1% of the total tissue.

Human cells were also detected in brains and livers. Most importantly, these cells were functioning. They were producing human proteins and remaining healthy months after birth. This could revolutionize organ transplantation by growing human organs inside animals, addressing the critical shortage affecting thousands of patients worldwide. Which I, I get it. You know, you hear the stories all the time.

People on these crazy waiting lists, not knowing if they're gonna get an organ in time [00:17:00] to be able to survive. So, absolutely. I get the why. But they said the human cells were detected in the brain . There are so many issues with that, and this raises huge ethical questions. Like where is the line between animal and human? Are we building new life forms or just solving a problem with very squishy and potentially doomsday consequences?

Like we are messing with creatures where one female can reproduce a hundred pups a year. Imagine if just one escapes and then somehow meets up with an escaped mammoth mice and decides to reproduce , like, what, what becomes, I don't, I don't wanna think about it.

I don't wanna live in a world where that's the case, so we're, just not gonna go there. But I will say this reminds me of James Patterson Zoo, which was also turned into a series. The first two seasons were great.

But I could not watch the third. In case you did not read the book or see it. I'll just give a quick summary. So I guess spoiler alert . So in Zoo animals started exhibiting strange hybrid like behaviors, lions attack and tourists, um, household pets, turning on owners and just really animals coming to the conclusion that humans are only at the top of the food chain because of weapons.[00:18:00] 

There's so much of them than there are of us. If they just came together with like, you know, the ants, the bees, the lion. Like they can take us out and they prove that they can take us out. So, um, and the cause of, why this happened, it's a, it's a variety of factors, including technology overreach.

What we see around us every day, AKA scientists, were manipulating genetics in ways that blurred the lines between species. Well, it's looking like fiction just became factual. I hope you guys are ready. So there you have it for my episode. Uh, was there any positive things in there?

No, there wasn't. I'll do better next time. Y'all, I'll do better. There are some that might think, ah, I think you're being dramatic. There's a lot of good in there. There probably is. There is, um, the, the last one with the mice growing organs, there is potential there, right? Growing organs so that people could get them in time and they have a chance at survival. I'm just, everything else about it, I. I am torn. When it [00:19:00] comes to science, it, it excites me and it also terrifies me, but I think it mostly terrifies me. I have, um, incredibly bad eyesight, so I'm hoping for a future where there are bionic eyes and I could see 2020. That would be awesome. I, I could get behind that, you know, as long as they're not growing whatever I need in mice, I'm cool with that. So yes, science excites me, and it also terrifies me. And I think that's a good thing, right?

We gotta consider the consequences, I don't wanna preach, and I gotta say we're only halfway done with 2025. Um, actually we're a little bit more than halfway done 'cause it's already July, but.

I'm interested in seeing what the rest of the year brings and even like at the rate things are heading, even the next two years. It's gonna be a wild ride. Sorry for that rambling. Uh, thanks for tuning in. If you like today's episode, make sure you subscribe so you don't miss what the future throws at us Next.

And leave a review, a comment, share with a fellow doomsday hobbyist, or just scream into the void with me at sci-fi realities. I'm also on YouTube, apple, all of the stuff where you find people that you know do [00:20:00] this podcasting thing. Until next time, y'all stay curious, stay cautious, and be very, very careful what you teach your AI. Today, it writes your lovely emails, but tomorrow, tomorrow it might write your exit strategy.