Imagine all the criminals,
It’s easy if you try,
With the power to do anything,
Enabled by AI.
The tremendous leaps being made by Artificial Intelligence Agents, or Agentic chains is phenomenal. It seems like every day we learn about a new way to use Agents to not only support workers, but to complete entire pieces of work.
In this article Ill take you through what Agentic Chains are and then flip the lens and look at how criminals, in an unregulated ecosystem could use them to plan or commit their crimes.
This is not a Chat-bot
To be clear, I’m not talking about chat bots, or those irritating tools that try so hard, but which have us quickly typing ‘connect me to a real human’ when they can’t answer our online query.
No, I’m talking about Agents, which are a subset of AI systems with massive promise. Let me give you some example prompts to demonstrate what exists now.
- Design a computer chip that does XYZ. Then create the digital files that can be sent to an integrated circuit manufacturer so I can test a prototype.
- Write me a complete set of computer code that does XYZ. Then create me an Android App that will XYZ.
- Design a rotor blade for a new drone. Test the physics and the materials. Then create me a 3D printer file so I can print my own at home.
- Design a new enzyme that will build a protein for me that I can use to create a new pharmaceutical.
- Create a new Large Language Model that will focus on topic XYZ so that it becomes an expert in that topic. Then create an interface that will allow me to upload a document, plan, or prompt so that your model can test it and identify any flaws in my plan.
Solving Problems Using Agentic Chains.
Agentic chains (or Hierarchical Agents as they are sometimes known,) now offer the capability to take complex tasks and break them down into phases.
Agentic chains don’t need human interaction at each decision point. They make decisions and act on them as they work through a complex problem. If the top-level agent can’t solve the problem, then it allocates tasks to one lower down in the hierarchy and so on until the problem is solved.
This creates a space for project manager type agents, which specialise in understanding the problem and working with the human to understand what is needed. Those project managers then solve the problem by outsourcing to model based, goal based or utility-based agents to name just a few. For the sake of this article, I’ll just call them all Agents.
The Growth of Agents.
Individual Agents are being created daily and are becoming more and more specialised. As they specialise, they need funding and so offer subscription payment models, or usage fees. That allows someone to effectively rent or use the system while not having to build and maintain their own.
Let’s take that a step further though. Imagine you have access to an Agentic Chain. It’s ultimately coordinated by a Large Language Model or LLM similar to a publicly available interface like ChatGPT or CoPilot.
You ask your interface to do something complicated that might involve hundreds or even thousands of steps. Your tool doesn’t have to do it all itself. That’s because it links to all the appropriate smaller Agents in the chain. In Agentic Chains each Agent solves a piece of the puzzle. Some Agents drive the process, and some design the pieces. Perhaps other Agents check the work, or run the finished product through physical simulations.
Criminal use of Agentic Chains
Let’s shake it up a little and think this through with a criminal lens. Stay with me while I describe something a little fictional to outline the thought process.
I log into my Tor Browser and find a criminal tool called, let’s say Blackforge. Next week I want to rob a bank. I describe what I want to do and upload photos, a building schematic, planning documents, the location, and any relevant details. I start off by asking Blackforge to create a plan for me, detail what tools I need, and where I can buy them.
Next, the tool looks at Police station locations and alarm response services. It works out what average response times are. Perhaps it even uses a telephony tool to call in a false report of criminal activity to actually test it. Or perhaps there is a tool that can bombard the emergency services with hoax calls on the night of my heist. Could I draw the responders away from my actual crime scene?
The next tool that Blackforge outsources to is a hardware hacking tool. Perhaps it can break into alarm systems, or video cameras, or even trip power grids in the neighbourhood.
And there’s more.
Another Agent has been trained on publicly available court transcripts. It searches for the most common causes or evidence on how criminals were caught in specific crime scenarios. It offers advice on what not to do, and therefore how to avoid getting caught.
Another Agent called ‘LawyerUp’ looks for lawyers with the best real world statistical outcomes on getting their clients acquitted. It produces a list of emergency telephone numbers.
Perhaps there is a tool that is a bit like a criminal version of Tinder! It matches you with available people that have specific skill sets which might be needed to pull off the heist. (I wont comment on trust, swiping, or whether they have a five-star peer review!)
And finally there could be an Agent that has learnt the most effective ways to launder your ill-gotten gains. Consider that it could be cash, gold, bitcoin or jewellery.
That’s probably stretched reality far enough for now, but you get my point.
Agentic Chains – Current Reality.
Whilst complex Agentic chains are currently only in the concept phase, Simple ones do already exist. As complex Agentic chains evolve there will obviously be money in them for the darker side of humanity and not just for business.
There will obviously be considerable trust issues, with historical sting operations like the FBI’s 2023 Trojan Shield success front of mind. If that can be resolved then there will of course be costs. At the moment, each large Agent or LLM is in the millions of dollars bracket, but that’s already come down from the tens and hundreds of millions. If you are only renting it for a few minutes, while it completes its task, then its considerably cheaper and already within reach. Potential success for a builder or owner of a criminal agent then narrows down to being able to create a successful brand perhaps along with anonymity. That’s obviously hard too.
That said, plenty of criminal organisations have worked quite happily on the Dark web for years. Granted, they have to keep reinventing themselves, change IP addresses and all the rest. That would make it harder for Agentic chains to function because the links would keep changing.
How Likely is this?
Currently, the majority of people don’t have the skill set or capability to make use of the Dark web or the Agents that may eventually be hosted there. Perhaps some grey zone agents also start to pop up, which can coach you through how to do it.
Alternatively, once an organised crime group gets to a certain scale, they don’t need to fund their model through ‘selling’ a service. They can build the tools inhouse, on owned architecture that is hidden suitably in hard-to-reach countries. Tools and models like that can evolve a long way away from any regulatory or investigatory jurisdictions.
And that of course is what could make them so dangerous.
Please subscribe and share.
I do hope you enjoyed this Reepaman Article, don’t forget to subscribe to receive them every couple of weeks. I would be grateful if you could also like or share as it helps me expand my reach.
Reepaman writes both fictional stories and truth based articles and looks at the criminal use of AI and drone technologies.
If you don’t yet know who Reepaman is, he’s a fictional character, and a psychopath. He could have been an Altman, a Bezos or a Musk, but he chose a different path, a darker path. Reepaman built an AI like no other, in secret, with no controls, no morals and no conscience.
The first Reepaman novel will be published in 2026.
#AgenticChain #AIAgent #AIAgenticChain #Reepaman #AI #Fiction #CrimeThriller #Booktok
If you want to learn more about AI Agents here are some great resources.