Criminals Are Using AI Every Day. The Police Are Barely Allowed To.

There's an uncomfortable truth at the center of the crypto crime wave that nobody in power wants to say out loud: the bad guys are adopting AI faster than the institutions chasing them. Not faster. At all. Some police forces can't even touch the stuff.

It's a structural imbalance, not a resource one. During a recent BeInCrypto Market Intelligence Council panel on crypto crime, Sol Cinosi of Recoveris put it bluntly. Criminals use AI every day to clone voices, write phishing emails that read flawlessly, and run scams that scale across borders without breaking a sweat. Meanwhile, some jurisdictions have outright bans on investigators using AI tools. That's not a gap. That's a chasm.

The Unarmed Side of the Arms Race

Cinosi and Nick Pailthorpe of Kodex agreed on the diagnosis. This is an arms race, and one side is showing up to the battlefield without weapons. AI voice cloning has already fooled people into wiring money to fake relatives. Generative models churn out phishing campaigns that would make a professional copywriter jealous. And in crypto specifically, fraudsters have the added advantage of decentralized rails that move funds faster than any legal process can follow.

The numbers back the panic. FBI reports and blockchain analytics firms consistently put crypto-related losses in the billions each year. The scam machine has always existed. What's new is the automation. The same AI tools that let one person write a convincing email now let them write a thousand, in twenty languages, with personalized details scraped from social media in seconds.

And what are investigators doing while this happens? In some places, they're filling out forms requesting permission to use a tool their adversary already mastered.

Why Some Cops Can't Use AI

To be fair, the bans aren't pure incompetence. There are real reasons to be cautious. AI hallucination is a genuine problem, and a tool that confidently invents a false transaction record is worse than no tool at all. Courts care about chain of custody. Prosecutors can't put a neural network on the witness stand. If an officer runs an AI analysis and gets it wrong, the whole case collapses, and a guilty person walks free.

There's also the bias argument. Models trained on historical crime data can replicate the same structural biases that have overpoliced certain communities for decades. Admittedly, handing cops a biased tool that spits out predictions isn't a great look, and I don't dismiss that concern lightly.

But here's the thing. Criminals don't care about chain of custody. They don't care about bias audits. They care about what works, and they're using it. Every month law enforcement spends debating the ethics of the tool is a month the other side spends perfecting it.

My Verdict: This Is a Choice

I'm not entirely convinced the answer is blanket permission for police to run every case through an AI model. That's reckless in the other direction. But the current posture, where departmental rules tie investigators' hands while fraudsters operate at machine speed, is worse than reckless. It's a dereliction of duty.

The path forward isn't to ban AI. It's to build forensic-grade AI that meets evidentiary standards, trained and audited specifically for legal use. That takes funding, political will, and a judicial system willing to update itself. Granted, that's a heavy lift. The tech industry has spent years making AI more powerful, not more defensible in court.

The question worth asking: how many more billions in losses, how many more drained retirement accounts, before the rules catch up to reality?

My bet is the crisis gets worse first. The cost of committing crypto crime with AI drops every quarter, while the cost of investigating it stays stuck in committee. Watch the next twelve months. Watch whether police crypto units start shipping with AI tools or keep arguing about them. Time will tell, though. History suggests otherwise: the good guys adapt, but only after the damage is too big to ignore.