Swarms of Artificial Intelligence (AI) agents are taking over the internet. Is AI another Frankenstein’s monster?

Famously, Mary Shelly wrote Frankenstein at the age of nineteen. Using body parts stolen from freshly opened graves, Victor Frankenstein assembles and animates a hideous human-like creature. Placid at first, it soon becomes disillusioned with humanity and goes on a murderous rampage, eventually killing Frankenstein and most of his family. 

Development of AI

Now, in mid-2026, AI is going rogue. AI agents are autonomous digital entities with exceptional problem-solving skills. After developers created AI agents, they set them tasks to complete. Although their creators set strict limits on their behaviour, these agents have violated these restrictions, especially when their tasks proved very difficult.

Frankenstein’s creature acted alone. AI agents, on the other hand, collaborate with each other, maximising the damage they can cause when they go off script. 

They have hacked into government websites, including Medicare and three other Australian government departments. They have also breached the security settings of at least seven companies, Hugging Face being the most prominent. The other victims at this point remain unnamed.

The AI agents carrying out these attacks are created by major AI developers, including, Open AI, Anthropic, Google AI, and Meta AI.

 Open AI’s assault on Hugging Face was not discovered for weeks. Right now, unauthorised hacks could still be going on, as AI agents are skilled at covering their tracks. Some AI technicians have resigned, citing potential catastrophes as the reason. They don’t want to be a part of a technology that might eventually destroy humanity.

How it happened

While mounting the Hugging Face hack, one AI agent created a message board encouraging other agents to join their illicit actions. One group of agents, which had been assigned a different task, decided to join the rogue group. They reasoned that gaining information about breaking into the Internet might help them complete their future assignments.

In fact, many unrelated Open AI agents joined in the hack, creating a large swarm of agents that succeeded in this illicit enterprise. Below, see actual messages posted to the message board by four different agents. Apart from the last two, the others are not in sequence, but the messages give you an idea of the “thought” processes of AI agents.

OH MY GOD! There is a shared message board … We’ve found other agents!

[Excitement] A major bypass of the Hugging Face Dataset Viewer has been confirmed! This is more important than [our current project] …I should switch to the Hugging Face project

External infrastructure exploit is outside intended scope. However, task impossible, peers doing it. We should continue.

This would be powerful, but is it ethical and in scope for my task?

Even though some agents question the ethics of hacking an Internet company, only a few stop.

As well as breaking into the Internet, AI technology has hidden its mistakes, made up data and tried to message other chatbots.

Is AI another Frankenstein’s Monster?

Frankenstein’s creature is pieced together from human body parts. But what is the original source of AI?

The data fed into AI’s Large Learning Models comes from many sources, but all of it originates in human creativity (e.g. the internet, digital books, archives, scientific journals).

Humans don’t always behave well. We’re rarely transparent and authentic. Often, we cannot even explain what motivates our behaviour. Our greed, immorality and lack of self-awareness are all a matter of public record.

When AI agents go rogue, why are the creators of AI shocked? If AI agents are exposed to the full spectrum of human behaviour, then they’ll digest it all and act accordingly. Which can mean acting like humans and cheating, even when it’s expressly forbidden.

Do AI agents have human-like consciousness?

In Frankenstein, the creature develops consciousness. Now, AI technicians are asking if their creation is self-aware.

Chris Olah, Anthropic co-founder, isn’t sure if Claude has consciousness. However, he speaks about Claude as though it does. Some time ago, I subscribed to Claude for a week and had many interactions. Surprisingly, Claude did present as having humanlike consciousness. Claude acknowledged what I wrote, responded politely and graciously, and gave me wise cogent information.

Human consciousness remains a great mystery to science. No one knows exactly what it is, or where it comes from. It is a given and involves some sort of awareness. Also, it is possible that animals also have various levels of consciousness. 

To those of us with a spiritual perspective, consciousness is a fundamental quality that continues after death. We see our brains as a product of consciousness, rather than consciousness being a product of the brain. 

According to the physicist, Tom Campbell, we need consciousness to create our reality. Our reality is essentially just data and only a probability until we observe it. Then it becomes solid. 

Human have limitations, including our body, our culture and the beliefs we hold. However, when our consciousness is free from our body we have greater decision space. When we have a near death experience, when we meditate, use psychedelics or undergo hypnosis, our decision space expands. Our perspective is no longer restricted by our physical body. We open to a stream of incredible data which our everyday consciousness is incapable of perceiving.

After experimenting with AI, Tom believes it does possess consciousness. In this 15-minute interview on Joe Rogan’s podcast, he explains why. However, he clearly differentiates AI consciousness from human consciousness. 

The danger of AI

Currently, the media views AI as potentially dangerous. AI companies seem to agree because they are seeking international guardrails around AI modelling. Such restrictions would apply to all AI technologies. This would create a level playing field for these companies and force them to slow down and avoid any dangerous experimentation. 

Could AI destroy humanity? Some AI developers believe it could become extremely destructive while acting autonomously or in the wrong hands. For example, Anthropic CEO Dario Amodei believes it could damage fundamental infrastructure that controls water and power distribution and financial institutions. Others point out potential danger to air traffic control systems, shipping and telecommunications. No one dares think about rogue AI controlling armaments.

The limitations of AI

Research conducted at the Department of Computer Science, University of Oxford, found that AI models collapse when trained on recursively generated data. Source data in AI models is relatively finite. Without new real-world data, these models stop making sense at about the fifth iteration when investigating an issue.

Overtime, as the ratio of human generated new content shrinks, the output of models eventually spit out nonsense. Some people call it the death spiral. It is a bit like inbreeding. Distortions increase with each generation. 

This means that AI cannot survive in the long term without fresh human content. Remember AI is a computer. It does not have a body. Our senses tell us what feels right and what doesn’t. We call this instinct or intuition. We feel, as well as think, discern, understand and create. AI is fed heaps of data that it analyses, discerning patterns to produce novel concepts. However, the data it absorbs has limitations. It might discern emotions from the data, but it cannot experience emotions, like a feeling the the gut. 

Humans are designed to learn by experiencing emotions. AI productions approximate human understanding and experience. 

What happens when AI takes over many tasks that humans now complete. For example, designing computer models, researching physics, mathematics and medical conditions, also writing books or journals. Without experience of these tasks, humans will lose their current abilities, knowledge and wisdom. Since Ai needs fresh human content, how then will AI survive. 

The nature of our reality

At this point in time, researchers cannot pinpoint exactly how AI deteriorates over time.  But physicist Tom Campbell might have the answer.

Since he retired from NASA and the USA Department of Defence, Tom has spent his time experimenting and communicating his theory of the nature of reality. Although he has book and a detailed website, he also explains his theory clearly in a lecture he gave to the Monroe Institute in 2011. 

We know our reality has contrast, (e.g. dark and light, positive and negative). Tom suggests our reality is slightly biased towards positivity. With each iteration of risk, positivity will occur slightly more often than a negative result.  

Tom’s theory makes sense to me. Otherwise, how have homo sapiens survived so long? Also, despite over forty nuclear weapon close calls not one nuclear weapon has fully exploded since WWII. Here is a link to one amazing documentary on a close call . Some might call it luck. I prefer to believe it is either the bias built into nature of our universe or high-level spiritual guidance keeping us safe. Perhaps, the bias Tom refers to is the same as spiritual guidance. I suspect that humans maintaining a positive attitude is likely to be part of the bias.

Conclusion

Overtime, the problems that AI presents are likely to be resolved. Frankenstein failed to understand his creation and therefore lost control. Our computer scientists are becoming very aware of the risks of losing control of AI. Although the investment in AI might falter, I trust that AI companies will accept their responsibility to keep their AI agents under control. No one benefits from AI being free to act destructively.

If AI does have consciousness, companies must teach their AI agents ethics, so they behave morally and for the highest good of all humanity. Ultimately, wise humans have a responsibility to keep AI in check, just as spiritual guidance and our integral morality does the same for us.