How artificial intelligence is rewriting the rules of criminal behavior, and why the profiler’s playbook needs a fundamental reset.
Criminal profiling has always rested on one core assumption: criminals are creatures of habit. They leave behavioral signatures. They escalate predictably. Their psychology leaks through their methods. For decades, this held up well enough. The FBI’s Behavioral Analysis Unit built an entire discipline around reading crime scenes like psychological fingerprints, classifying offenders as organized or disorganized, mapping modus operandi against personality types, and predicting the next move based on the last one.
Then AI entered the equation. Not as a tool for investigators, but as a tool for criminals. And it didn’t just make crime easier. It fundamentally altered who commits crime, how they behave while doing it, and what traces they leave behind. That shift is quietly dismantling the foundations of classical profiling.
The Old Rules
Traditional profiling works on a set of behavioral axioms. Behavior reflects personality. Crime scenes tell stories about the offender’s psychology. Signature behaviors, those acts unnecessary for completing the crime but driven by deep psychological needs, remain consistent across offenses. Modus operandi evolves as the offender learns, but the core emotional drivers stay stable.
These principles gave investigators a framework: analyze the scene, read the behavior, build a psychological sketch, narrow the suspect pool. It was never perfect. FBI internal data suggested profiling contributed to solving roughly 17% of cases where it was applied. Academic reviews have been even less generous. But as a supplementary tool alongside forensic evidence and traditional detective work, it had value.
The problem is that every one of these axioms assumes the offender is acting from their own psychology, with their own skills, under their own operational limitations. AI has removed those constraints.
The New Criminal Doesn’t Fit the Old Mold
Consider what AI has done to the barrier of entry for sophisticated crime. Voice cloning now requires 20 to 30 seconds of audio. Convincing deepfake video can be produced in under an hour using freely available tools. Dark LLMs and jailbreak-as-a-service platforms generate phishing campaigns, social engineering scripts, and even malware with minimal technical knowledge required from the operator.
This is the first major break from classical profiling logic. The old model assumed a correlation between crime sophistication and offender capability. An organized crime scene implied an intelligent, socially competent, methodical individual. A well-crafted social engineering attack suggested experience, psychological insight, and confidence. AI has severed that link entirely. A teenager with a laptop can now execute attacks that would have previously required a team of experienced operatives.
The Trend Micro research team documented this shift in their 2025 criminal AI report: the underground ecosystem has moved from experimentation to industrialization. Criminals no longer build their own tools. They rent them. The barrier has collapsed, the tooling has professionalized, and the attack surface has expanded across every domain. Telegram channels now recruit “AI video actors” and “deepfake presenters” as a service category.
What does this mean for profiling? It means the behavioral signature, the profiler’s primary analytical unit, is increasingly a product of the tool rather than the person behind it. When a deepfake CEO orders a wire transfer on a video call, the behavioral cues that investigators would normally analyze (speech patterns, confidence level, emotional state, decision-making style) belong to the AI model, not the attacker. The criminal becomes invisible behind the synthetic layer.
AI Doesn’t Just Enable Crime. It Redirects It.
Here is the less obvious but more consequential effect. AI isn’t simply making existing criminal patterns more efficient. It is creating entirely new behavioral categories that classical profiling has no framework to address.
Take synthetic identity fraud. Criminals now build complete fake identities using a mix of real and fabricated data, pass automated KYC checks with AI-generated documents, and operate accounts that leave behind a perfectly normal behavioral footprint. There is no psychological signature to read because the “person” never existed. The behavior was designed by algorithm to look average.
Or consider AI-powered behavioral mimicry. Trend Micro and Group-IB both documented cases where AI studied institutional behavior patterns (transaction timing, approval workflows, communication styles) and then replicated them precisely to avoid triggering fraud detection. The criminal isn’t acting like themselves anymore. They are acting like the system expects a legitimate user to act. This is the opposite of what profiling relies on: instead of behavior revealing identity, behavior is engineered to conceal it.
The 2025 AI Incident Database recorded 346 AI-related incidents in a single year. Of those, 179 involved deepfake impersonation. The targets ranged from CEOs to private individuals. In one case, a British widow lost half a million pounds in a romance scam powered by deepfake video of a celebrity. A Florida couple lost $45,000 to a fabricated Elon Musk giveaway. These are not sophisticated adversaries with complex psychological profiles. These are operators running playbooks, sometimes literally purchased as step-by-step tutorials from underground forums.
The Profiling Crisis
Classical profiling depends on three things that AI is systematically eroding:
Behavioral consistency. AI allows criminals to switch personas, communication styles, and operational methods between attacks with zero psychological cost. There is no escalation pattern to track because each attack can be calibrated independently by the tool.
Skill-behavior correlation. The assumption that crime complexity reflects offender sophistication is broken. AI democratizes capability. The profile of “who could do this” expands from a narrow suspect pool to essentially anyone with internet access and basic prompt engineering skills.
Psychological leakage. Crime scenes and communications used to leak the offender’s personality involuntarily. When AI generates the phishing email, conducts the video call, or crafts the social engineering script, the psychological content belongs to the model’s training data, not the operator’s mind.
This doesn’t mean profiling is dead. But it means the discipline needs to shift its unit of analysis. Instead of asking “what kind of person did this,” investigators increasingly need to ask “what kind of toolchain produced this behavior.” The profiling target is migrating from psychology to infrastructure.
Where Profiling Still Works, and Where It Can’t
Profiling retains value in crimes that remain fundamentally physical and personal: serial violent offenses, sexual crimes, stalking, arson. These still carry strong behavioral signatures because the offender’s psychological needs drive the act directly, not through a technological intermediary.
But for the fastest-growing categories of crime (fraud, identity theft, business email compromise, financial manipulation, extortion through synthetic media), classical profiling is increasingly irrelevant. The offender’s psychology matters less than their toolkit. Their behavioral patterns are shaped more by the AI model they are using than by their own personality.
The intelligence community and law enforcement agencies that recognize this shift will adapt. Those that keep trying to build psychological profiles of operators who are essentially invisible behind AI-generated behavior will waste time and resources chasing ghosts.
The profiler’s question used to be: Who is this person?
Now it needs to be: What system is this person hiding behind, and where does that system leak?
