Reepaman – 6 – The Artificial Criminal

Reepaman works with AI to commit crime

The power of AI is terrifying – in the wrong hands.

Reepaman was sat in his den, surrounded by bits of half-built drones, 3D printers and computer screens. “Thalia. We need to start a new project. I want you to create a new Agent and data set.”

“Alright Reepaman. Are you building it on my server or shall we rent discrete space from the Hungarians again?”

“Good question. Let’s go with the contractors. We need an air gap between us for this one.”

“No problem, what do you want to call the Agent?”

I thought for a moment. What the heck do you call an Agent that you want to build, that’s going to have the sole purpose of planning how to assassinate someone?

“Leon,” I said.

“And what’s the core role of the Agent?”

“Thalia. I want us to build an Agent that is going to help me plan and conduct an assassination. Needless to say, this is for my knowledge only.”

“Oh, that’s sounds interesting. Do you mean murder, or assassination?”

I pondered for a moment. “This is definitely a public figure, so assassination, though I wouldn’t rule out using the tool for other killings in the future.”

“I get the Leon reference now. It’s the French assassin movie with Jean Reno and Nathalie Portman.”

“Yeah. Now what key research areas and tools do we need for Leon to be an assassins professional guide?”

You’ve probably guessed by now, but Thalia is my own personal AI tool. She’s a little rough around the edges, but is very useful for me to plan, design and build the tools I use. When I’d been at MIT my girlfriend and I’d worked out how to partition 0.01% of MIT’s super-computer and had corrupted it for our own personal use. My girlfriends no longer around but Thalia’s never been detected, and all of her activity is lost in the noise of all the research being done there.

Reepaman works with AI to commit crime
Reepaman: Created using ChatGPT

The primary benefit, other than running a small AI system for free was that I could pretty much ask Thalia any question I wanted to with out raising the flags that you and I would normally raise if our web browsing activities started to get too weird. Trust me, I’d have been on more watchlists than a TikTok conspiracy theorist during a G7 summit.

Thalia brought up a list on my laptop screen while her voice continued over the speakers in my den.

“We need to break this down into manageable pieces Reepaman. Can I suggest that I give you the headlines and then I try to digest this overnight to work out the details?”

“OK, go for it.” I reached for my can of Coke and took a swig.

“We need to understand the target, who they are and what they do in terms of behaviours. What are their routines, and habits? What security measures are in place to protect them, either at home or in public. Ultimately as an individual, what are their vulnerabilities? We can use a number of tools there to help us. Social media of course, digital footprint, phone records, biometrics, hell, even their fitness app can tell us if they run the same route every Tuesday morning before work.”

“Go on.”

“Aside from social media, there are lots of public records. Home addresses, vehicle records, company and directorship records. We can use a lot of public media too for high profile targets.”

“Like what?” I asked.

“Well, if you are somebody, and you’re going to speak at a conference, you can bet the conference organisers will publish the date and time of your speech. If they’ve paid you an appearance fee, they definitely want to get the paying public in the door. If you are a big enough personality, the local news will be all over it too before the event. Then you get fans posting about having seen you in their local coffee shop while they tag your social media account. All of that is traceable and can build a very solid picture of your habits and routines. You might be a private person, with all your own social media restricted or managed, but it’s everyone else that creates the intelligence picture about you.”

I was pretty sure I knew most of this, but it was scary when it was all joined up. Luckily, I wasn’t famous, and I was absolutely fanatical about my own data footprint.

Thalia continued, “Once you know your target inside out, you need to expand the picture. Based on what you know about the target and their behaviours, start reviewing the potential assassination locations. Consider them your battlegrounds because you need to understand the location intimately and control it to your advantage. What’s your route in and out. What’s the emergency exit if things go wrong? Where are the surveillance zones and are there any blind spots? What are the law enforcement response times and how will they get to the scene?” She paused before going on.

“Next, weapons. How will you conduct the hit? There are a million choices from poison to firearms, or an overdose to an automobile. You also need to decide if you want to make it a blatantly obvious statement event, or try to hide it as an accident or a medical problem. Are you going to try to increase the odds of guaranteeing the outcome, but assume a higher risk of getting caught? Or, do you want to better the odds of getting away, but reduce the likelihood of success of achieving the aim. There are so many variables here that we could probably fill an entire database on that alone.”

Reepaman works with AI to commit crime
Reepaman: Created using ChatGPT

Thalia certainly had a point there and I said, “Let’s assume that I need to complete the task undetected and with the priority being getting away with it.”

“Alright, so that will reduce the chance of success.”

“Understood. What’s next?”

“The next heading is ‘Deception.’ Do we want to deceive the victim into doing or not doing something so that we can increase their vulnerability? Or, do we want to deceive or obfuscate the security forces when they respond? Are we using any social engineering techniques, deepfakes, or social media cues to shape events or create confusion? Alternatively, are we going for complete simplicity, reducing the complexity and therefore the chance of any one part of the plan going wrong.”

“That sounds good to me.”

“Ahh, but with a target with very few vulnerabilities and a great security team, all the easy approaches will have already been planned for and countered. Sometimes things have to be complicated, but that of course increases the resources needed.”

“Thalia, I’m not trying to assassinate a President here, just a normal human being.”

“If you say so Reepaman. Next up is personal security and counter-surveillance. You and I already do a lot of that, so we don’t need to build anything new there.”

“OK, but don’t discount it. Double check our approach.”

“OK. Planning and workflow automation is the next heading.”

“Workflow automation? Seriously? We’re talking about criminal activity here, not a production line.”

“I know, but to do things right you need a plan, and in that plan, you want to trigger actions at the best moment. Some of that can be automated.”

“Like what?” I asked.

“Let’s say NBC news is live broadcasting the arrival of your target. They show on video the target arriving and their airplane door opening. You might automate the launch of a drone swarm at that point, or perhaps start a pre-planned social media campaign or other diversion. By Automating some tasks which are pre-planned, you can reduce the number of people involved, and so improve project security and speed.”

“Fair enough. It sounds complicated again though.”

“Ahh, but isn’t that the point Reepaman? You want to build an AI tool that is going to be effective while guaranteeing success and escape. Sometimes things will have to get complicated, and I can help you plan them out in detail. We can also come up with contingency plans for a whole range of scenarios where things might go wrong.”

“Okay, Okay, I get it. So, with all these planning stages, what data sets do you need?”

“It’s easier if I put them up on your screen and then you can ask key questions.”

Thalia proceeded to list the data sources on my laptop screen.

  1. Surveillance and CCTV Feeds (public or hacked)
    • To track target movement or identify security patterns.
  2. Social Media & Online Behaviour Data
    • Facebook, Twitter/X, LinkedIn, Instagram metadata.
    • Comments, habits, followers, photo geotags.
  3. Geospatial & City Planning Data
    • Building layouts, emergency exits, traffic patterns, camera locations.
  4. Biometric & Medical Records (if obtainable)
    • Blood types, allergies, implants, medications, health conditions.
  5. Criminal and Law Enforcement Databases
    • Arrest records, known associates, active investigations.
  6. IoT & Smart Home Device Logs
    • Target’s smart speaker, security system, car telemetry, fitness tracker, phone movement logs.
  7. Public Transport and Rideshare Logs
    • To track or predict movement (Uber, Lyft, metro swipes).
  8. Online Purchases & Financial Transactions
    • Buying patterns, travel tickets, time of day for purchases, insurance records.
  9. Dark Web Feeds & Marketplaces
    • For procuring tools, sub-contracting work, identifying competitors, or analysing chatter.
  10. AI-Specific Training Sets
  • Facial recognition, gait analysis, voice mimicry, and anomaly detection.

I’d read the list. “Shoot, can you actually get all those?”

“No, not without an element of hacking, but we can buy some of the software tools to hack some of those on the dark web. We won’t get any access to some of the others though, unless perhaps we coerce or blackmail an insider.”

“What about point ten? The facial recognition piece?”

“That’s actually very easy depending on who your target is. For a public person there will be an almost unlimited choice of photos available online. We can create a fully functioning 3D model, or deepfake video of pretty much anyone on the planet. Ahh, Reepaman, I just checked my work, there’s another key data set and tool that we need to build if we are going to do this properly.”

“Really? It feels like we have a lot already. I’m worried about the computing resources and data storage.”

“Yes, it could be quite large. But, we really want a Red Team agent.”

“Explain.”

Reepaman works with AI to commit crime
Reepaman: Created using ChatGPT

“You want an Agent, playing the opposition, that is going to critically look at your plan and all your data and monitor your progress as you implement it. You want something that is going to start alerting you if you are making a mistake.”

“Like what?”

“Well, you want an Agent that’s been trained on every crime ever committed. An Agent that is going to look at how the criminals were caught. Statistically what mistakes got them caught and how likely any particular course of action might lead you to getting caught during your crime.”

“Holy cow, that’s awesome. Then, we can tweak the plan or offer solutions to mitigate those risks. Can we actually do that?”

“I think so. We can start with all of the publicly available court transcripts for murder and serious crimes. We can add media reports, confession transcripts, and any other publicly available records and statistics. Also, let’s review all academic criminology papers too. Even the FBI publish how they solved some of their most serious cases. Imagine that data, from all around the world. There must be hundreds of thousands of high-quality data records out there, if not millions. And they are all free.”

I paused for a moment, as I thought about that. “Thalia has anyone ever done that before?”

“Not that I know of. Certainly, criminal groups don’t have the resources to do it. Or rather, they didn’t until tools like me were invented.”

“Thalia, add a piece of work to your work queue. I want you to do a business plan on selling access to a tool like that. If we were to market it through the dark web, could we sell its services and make meaningful money? Think of it like giving a criminal unrestricted access to ask questions about how to commit a crime. But, you know, without the irritating messages from publicly available AI’s that block you, ban you or report you to the government if you start asking really weird stuff. Let me know the pro’s and con’s after you’ve done some analytics.”

“Will do Reepaman.”

It was getting late, and my brain was getting a little fuzzy. I packed up for the day, with just a couple of thoughts running around my head. If I built this tool, I could probably commit any crime that I ever wanted, learning from the mistakes of all those who had gotten caught before me. If I built this tool, and offered paid access to it, crime rates around the world would surge, and I would make a lot of money.

Perhaps at that point I might design an AI for the good guys. But then again. Where would the fun be in that?

———————————

If you enjoyed this short story, and would like to get updates then you can follow my newsletter on LinkedIn where I regularly publish more Reepaman Stories. https://www.linkedin.com/newsletters/6919145812560150528/

Alternatively, sign up for the emailing list here: https://reepaman.com/index.php/contact-us/

Alternatively, it’s all free and available on the authors website www.reepaman.com.

This is a work of fiction. All characters, scenarios and events are imaginary, and any coincidences are unintended.

Having said that.

The technology to execute the events and systems described is here, now.

There are those with the capability and intent to use it…

My written work is original and not written by AI. I do use AI for research and to create the images that accompany my work. Personally, I find #ChatGPT to be a very effective tool for both writing the prompts to create the images, and creating the images as well.

#ReepamanChapter6 #FanFiction #Leon #Agent #Crime #AICrime #AI #ArtificialIntelligence #AI4Bad #AIWorstCase

#MilitaryThriller #TechnoThriller #Fiction #ShortStories #Free2Read #RobPhayre

© Copyright worldwide. Rob Phayre Ltd.

Images produced using #ChatGPT