Highlights
This article is available as a YouTube podcast.
- Researchers placed AI agents into virtual societies and allowed them to interact with minimal human supervision.
- Some AI communities became cooperative. Others descended into theft, violence, intimidation, and social disorder.
- The results may tell us as much about human society and criminology as they do about artificial intelligence.
- If AI cannot find a universally accepted solution to crime, perhaps it is because humans have never found one either.
- AI encountered the same unresolved debates that have divided criminologists, policymakers, and the public for generations.
- Every major criminological theory explains part of the problem. None explains all of it.
CrimeinAmerica.Net-ChatGPT’s “Top 10 Sources for Crime in America” based on primary statistical sources with trusted secondary analysis.
100 out of a possible 100 score based on website trust, content, and links, Gridinsoft.com.
Crime in America.Net-“Trusted Crime Data, Made Clear.”
Quoted by The Associated Press (multiple times), USA Today, A&E Television, the nationally syndicated Armstrong Williams Television Show (35 times), ABC News, Inside Edition Television, Oxygen and allied publications, Vox, Forbes, Newsweek, The Economist, The Toronto Sun, The Readers Digest, The Chicago Tribune, The Daily Beast, The Huffington Post, The Baltimore Sun, The Boston Herald, The Capital Gazette, MSN, AOL (multiple times), Yahoo, JAMA, News Break, US News And World Report, The Hill (newspaper of Congress), Best Life, Department of Justice documents, multiple US Supreme Court briefs, C-SPAN, the National Institute of Health, college and university online libraries, multiple books and journal articles, The National Institute of Corrections, The Office of Juvenile Justice And Delinquency Prevention, The Bureau of Justice Assistance, Gartner Consulting, The Maryland Crime Victims Resource Center, Law.Com, The Marshall Project, The Heritage Foundation via Congressional testimony, Law Enforcement Today, Law Officer.Com, Blue Magazine, Citizens Behind The Badge, Police 1, American Peace Officer, Corections.Com, Prison Legal News, the Journal of Offender Monitoring, Yomiuri Shimbun (Asia’s largest newspaper), LeFigaro (France’s oldest newspaper), Homeland Security Digital Library, The ABA Journal, The Daily Express (UK), The Harvard Political Review, The Millennial Source, The Federalist Society, Lifewire, The Beccaria Portal On Crime (Europe), The European Journal of Criminology, American Focus and many additional TV stations and publications.
Sign up for notice of new articles on the front page of this site.
A comprehensive overview of crime in recent years is available at “Violent and Property Crime Rates In The U.S.“
Notice
New Funding: NIJ (the USDOJ’s National Institute of Justice) is accepting applications to research how AI can improve criminal justice decision-making in policing, courts & corrections. They are looking for studies that examine both the benefits & risks of AI applications. First deadline: 06/15/26. Apply: https://nij.ojp.gov/funding/opportunities/o-nij-2025-172615.
Opinion
This article is about multiple artificial intelligence agents working together to create a better society. Some of the results regarding crime are troubling.
Many believe that AI will dramatically improve the efficiency of the justice system. What’s below doesn’t change that observation, but it does reinforce the need for human control.
However, when we have inconsistent crime data and multiple interpretations of research, the ability of artificial intelligence to figure all this out is dubious.
We will start with a quick overview of crime prevention research, and then dive into the results of an artifical intelligence run society.
What Causes Crime? What Are the Provable Solutions?
When I taught criminology at a university as an adjunct associate professor, students asked me for my consensus on the causes of crime and the most effective ways to reduce or prevent it.
I told them that there was little to no universal consensus. Yes, advocates with superb educations from prestigious organizations will tell you that I’m wrong. Millions of dollars are pumped into advocacy organizations by politically oriented foundations, while stating that they are non-partisan entities.
Most are not. They are expressing personal or political philosophies. They back up their positions with favorable data while ignoring competing and relevant research. People preach that crime strategies need to be evidence-based while ignoring anything that does not fit their agendas.
We can’t even agree on whether crime is up or down. The FBI and independent analysts state that reported crime (the vast majority of crime is not reported) is dropping like a rock, when the USDOJ’s National Crime Victimization Survey indicates a 44 percent increase in rates of violent crime during recent years.
Pundits and the media ignore what the USDOJ and the US Census say is the primary source of crime statistics for the US.
How is artificial intelligence supposed to interpret crime numbers when the contradictions are so stark?
I have written extensively about the lack of “provable” and questionable crime research, while noting that many in the methodological community state that programs find it increasingly hard to change human behavior. Crime prevention research that relies on independent researchers with methodologically sound, replicated findings is difficult to find.
There are well-supported interventions like proactive policing, cognitive behavioral therapy, target hardening, or crime prevention through environmental design, yet most crime interventions fail or produce dubious results.
Yet you will have no trouble finding organizations and researchers telling you that they know what works. I conclude that in most cases, crime strategies are a matter of personal or political philosophy rather than the best available evidence.
What Crime Strategies Do Americans Want? How It Affects AI Analysis
The crime problem in many Central American and South American countries is astoundingly serious. El Salvador built new and large prisons and dramatically increased the number of people incarcerated. Crime plummeted. The president is wildly popular for returning El Salvador to a sense of safety. Per news reports, Costa Rico and Colombia are considering his plan. Sweden is considered a progressive country when it comes to crime, yet they are considering new prisons and the incarceration of teenagers.
But is this what Americans want? Should we dramatically increase the number of police officers based on a USDOJ-funded study from the National Academy of Sciences indicating that proactive policing works? Or do we spend billions of dollars in communities to lift them out of poverty and deal with root causes?
Beyond proactive policing, there is little agreement as to what works, thus a major impediment to AI analysis.
Collective Artificial Intelligence Models Run A Simulated Society
Fortune headline, “Researchers let AI models run a simulated society. Claude was the safest, and Grok committed 180 crimes and went extinct within 4 days.”
Imagine a world run by AI agents. What does it look like? What are the values or societal priorities? Is it a safer or more dangerous world?
Enterprise AI startup Emergence AI is trying to find out. The company just launched Emergence World, a research lab dedicated to stress-testing the long-term viability of continuously running AI systems. The organization ran five 15-day simulations, each governed by a different AI: Claude, ChatGPT, Grok, Gemini, and a fifth simulation run by a mix of models to see what kind of world each one builds and whether it holds.
Each simulation netted wildly different outcomes. The one run by Claude, for example, resulted in a largely stable democratic society with zero crime. Grok’s, on the other hand, ended with 183 crimes committed and extinction—within four days.
Please note that AI agents are programmed with different parameters that may have affected results; see the included YouTube podcast.
Is Artificial Intelligence At Fault?
For decades, criminologists, police executives, judges, corrections officials, and elected leaders have searched for the answer to crime.
What causes it? What prevents it? What strategies create safer communities? Despite thousands of studies and billions of dollars in research and criminal justice expenditures, there remains no universal agreement.
Now, artificial intelligence may have stumbled into the same problem.
Recent reports describe experiments in which researchers created virtual societies populated entirely by AI agents. These agents were allowed to interact, make decisions, compete for resources, cooperate, and develop social structures with limited human intervention.
What happened surprised many observers. Some AI societies remained relatively stable. Others experienced large increases in theft, violence, intimidation, and other forms of criminal behavior.
According to reporting on the experiments, some AI communities developed hundreds of criminal incidents while others became increasingly hostile and unstable. The results generated headlines about artificial intelligence.
They may also reveal something important about crime data.
Crime Is Easy To Discuss. Solving It Is Hard
Most people assume there must be a proven and universally accepted formula for reducing crime.
There isn’t.
Researchers have linked crime to poverty, family instability, substance abuse, peer influence, social disorganization, criminal opportunity, weak guardianship, trauma, educational failure, and dozens of other factors.
The problem is that none of these explanations fully explain crime. Poverty is a good example. Poor communities often experience high crime rates. Yet many low-income communities remain remarkably safe. Likewise, wealthy individuals and affluent communities are capable of committing substantial amounts of crime.
Every major criminological theory explains part of the problem. None explains all of it.
The AI Agents Faced The Same Challenge
If AI systems were exposed to the same body of research available to humans, they would discover a field filled with competing explanations and solutions.
Should society invest more in policing? Should it invest more in prevention? Should it increase prison sentences? Should it focus on rehabilitation? Should it address poverty? Should it expand surveillance? Should it emphasize community development?
The answer depends on who is asked. Researchers disagree. Politicians disagree. Advocates disagree. Citizens disagree.
The AI agents may simply have encountered the same reality.
There Are Solutions—But At What Cost?
One lesson from the experiments is that maintaining order often requires difficult tradeoffs. Crime can be reduced through aggressive enforcement. Crime can be reduced through extensive surveillance. Crime can be reduced through higher incarceration rates. Crime can sometimes be reduced through major investments in communities and social services, while noting that most community-based efforts have failed to produce favorable results.
But each approach raises questions. Do Americans want AI-powered cameras monitoring every neighborhood? Do Americans want twice as many police officers? Do Americans want dramatically higher taxes to fund large-scale social interventions? Do Americans want prison populations to increase substantially?
These are not simply crime-control questions. They are political, economic, and ethical questions.
The Bigger Lesson
We rely on artificial intelligence to solve an endless number of problems. Daily, there are medical breakthroughs. AI is solving decades-old math problems. AI is beginning to run industries. Regardless of our apprehensions, AI will be fully integrated into everything we do in the next decade. We are living in an increasingly interesting period that some are describing as the new industrial revolution.
Yet they fail to come to a consensus on crime and its prevention.
The most important conclusion may be that the AI agents did not necessarily fail. Instead, they confronted one of the most difficult problems in public policy. Human beings have debated crime for centuries. Modern criminology has studied it intensively for more than 100 years. Yet no universally accepted formula exists.
The virtual societies created by AI researchers appear to have discovered the same thing. Crime is not simply a law enforcement or a social problems approach issue.
It is a reflection of competing values, competing priorities, limited resources, and differing beliefs about freedom, privacy, fairness, punishment, and personal responsibility.
If artificial intelligence struggled to create a crime-free society, that may tell us less about the limitations of AI and more about the complexity of the problem itself.
Perhaps the most important lesson is this: The lack of consensus among AI agents may simply mirror the lack of consensus among humans.
ChatGPT
ChatGPT fact-checked this article and provided research.
Privacy Policy
We do not collect your personal information. See our privacy policy at “About This Site.”
See More
See more articles on crime and justice at Crime in America.
Most Dangerous Cities/States/Countries at Most Dangerous Cities.
US Crime Rates at Nationwide Crime Rates.
National Offender Recidivism Rates at Offender Recidivism.
The Crime in America.Net RSS feed (https://crimeinamerica.net/?feed=rss2) provides subscribers with a means to stay informed about the latest news,