How Law Enforcement Uses AI to Fight AI-Enabled Crime

Law enforcement uses AI to fight AI-enabled crime the same way bad actors use it to commit crime: to do in minutes what used to take a team weeks. Investigators apply AI to triage digital evidence; resolve identities across data sources; link cases that, on the surface, seem unrelated; follow illicit funds across financial rails; and flag synthetic media. 

The agencies getting results are also using AI to connect evidence across disparate sources faster, with every step documented well enough to hold up in court. This is particularly important because AI-enabled crime rarely leaves its evidence in one place. For example, a single deepfake investment scam can involve a cloned voice, a rented phishing kit,  conversation in a Telegram channel, funds moved through a shell company, and a stablecoin cash-out. Tools that see only one of those layers see only part of the crime, making AI-enabled all-source investigative tools even more valuable as AI adoption spreads further throughout the criminal ecosystem.

Key takeaways

• AI-enabled crime is growing fast. The FBI’s Internet Crime Complaint Center logged nearly USD 893 million in losses tied to AI in 2025, the first year it tracked the category.
• Scams and fraud are the most mature AI-enabled crime type. TRM’s 2026 AI-in-Crime Adoption Index found reported deepfake-scam losses so far in 2026 already exceed all of 2025 by 263%.
• Police use AI to review evidence, resolve identities, link cases, trace money, and detect synthetic media. The biggest gains come from connecting data sources, not from any single tool.
• AI can help detect deepfake scams, but detectors lose much of their accuracy on real-world fakes. Investigators get further by following the evidence criminals can’t fake, like payment trails and reused infrastructure.
• Law enforcement AI has to be defensible: documented provenance, a repeatable method, and a human investigator exercising judgment and making the calls.

‍

{{horizontal-line}}

What AI-enabled crime looks like in 2026

AI-enabled crime is any criminal activity in which artificial intelligence (AI) is used to plan, execute, scale, or conceal the offense. In practice, that ranges from a scammer using a chatbot to write better messages to a full criminal service built around AI.

The numbers show how quickly AI-enabled crime has moved from theory to casework. In its 2025 Internet Crime Report, the FBI recorded 22,364 complaints and nearly USD 893 million in losses tied to AI, out of nearly USD 21 billion in total reported losses. TRM’s 2026 AI-in-Crime Adoption Index, which scores how deeply criminals have adopted AI across crime types, rose from 28 in 2024 to 54 in 2026. Scams and fraud were the only category rated “Mature,” with AI now present across the full operation, from targeting victims to laundering proceeds.

‍

{{59-how-law-enforcement-uses-ai-to-fight-ai-enabled-crime-callout-1}}

Why AI-enabled crime is an all-source problem

AI creates scale. It enables a single operator to run hundreds of conversations at once, generate convincing fake identities on demand, and use tools that once took deep technical skill. The result is crime that spans identity, infrastructure, and money simultaneously — and that can be deployed en masse in minutes.

As AI continues to expand the surface area of criminal activity, investigators are increasingly talking about all-source investigations: combining every relevant category of information — including open-source intelligence (OSINT), cyber threat intelligence, corporate records, sanctions lists, dark web intelligence, and blockchain intelligence — into one picture. The method comes from intelligence community practice, where no single source decides the outcome and conflicts between sources are treated as findings.

The crux of the problem isn’t a shortage of data, but rather, time, specialist expertise, and the manual work of reconciling one system’s output with another’s. This is where AI earns its place in an investigator’s toolkit.

Six ways law enforcement uses AI to fight AI-enabled crime

1. Triaging digital evidence at scale

Modern cases generate enormous volumes of phone extractions, video, chat logs, and documents. AI transcribes, translates, summarizes, and flags relevant material so investigators start with what matters. In the UK, early trials cited by the government’s new PoliceAI center included reviewing 800 hours of footage from a kidnapping case in three hours.

2. Resolving identities across data sources

Criminals using AI spread their footprint across usernames, email addresses, phone numbers, wallets, and companies. Entity resolution uses AI to recognize when scattered identifiers belong to the same person or network. An investigator who starts with a single wallet or handle can see that identifier’s links to social accounts, dark web forum activity, corporate registrations, and onchain activity in one session, rather than running a separate search in each system.

3. Linking cases that look unrelated

AI-enabled scams are industrial. The same operators reuse scripts, infrastructure, and payment addresses across thousands of victims in multiple jurisdictions. AI helps cluster victim reports and surface shared identifiers, so a local fraud report can be connected to a larger network. Scam-reporting platforms such as Chainabuse give investigators structured victim data to work from.

4. Following the money across financial rails

Most AI-enabled crime is financially motivated, and the money has to move. AI now compresses work like wallet clustering from days into minutes and lets investigators query transaction data in plain language. Tools such as TRM Forensics trace funds across blockchains to exchanges, over-the-counter desks, and other cash-out points where they can be frozen or seized.

5. Detecting synthetic media and identity fraud

AI detection tools analyze images, video, and audio for signs of manipulation, and help financial institutions spot synthetic identities at onboarding. These tools are useful, but they are one layer of the response.

6. Disrupting criminal infrastructure

The most effective responses to AI-enabled crime don’t stop at one arrest. They take down the services and infrastructure that let criminals scale. The EvilTokens action combined civil litigation, infrastructure takedowns, financial tracing, and arrests. The FBI’s Operation Level Up takes a preventive approach, using intelligence to notify people who are being scammed while it’s happening. According to the FBI, it has reached more than 8,000 victims and reduced losses by more than USD 500 million since it began.

What AI tools do police use to investigate crimes?

Police and investigators use AI across six broad categories of tools. Which ones an agency uses depends on its mission, budget, and policies.

Digital evidence and media review

Tools that transcribe interviews and calls, translate foreign-language material, summarize documents, and flag relevant content in phone extractions and video. These are among the most widely adopted, because they save time without making decisions.

Report writing and administrative assistants

Generative AI tools that draft police reports from body-camera audio or notes. The US Department of Justice COPS Office has examined how agencies are using them, and how officers must still review and own what the report says.

Facial recognition and license plate recognition

Facial recognition compares images against databases to generate investigative leads, and automated license plate readers log vehicle locations. These are also the most scrutinized. The Department of Justice’s 2024 report on AI and criminal justice recommends that facial recognition results be treated as a lead, not as sufficient to establish probable cause, and that trained examiners review matches.

Open-source and social media intelligence

Tools that collect and analyze public web content, social media, and forums. Social media intelligence (SOCMINT) and dark web monitoring help investigators attribute online personas and spot emerging threats.

Blockchain intelligence

Platforms that attribute crypto addresses to real-world entities and trace funds across chains. Because so much AI-enabled fraud and cybercrime is paid for and cashed out in crypto, blockchain intelligence is often where a case moves from a username to a subpoena.

AI investigation platforms

The newest category: platforms that combine many of the capabilities above, running a single investigative question across multiple data sources at once. The strongest of these keep the investigator in control, show how each conclusion was reached, and produce output that can be documented or reproduced for a prosecutor.

Can AI detect deepfake scams?

Yes, partly. AI detection tools can flag many manipulated images, videos, and voice recordings, but they are far less reliable on real-world fakes than lab results suggest. A 2025 study of deepfakes circulating online, Deepfake-Eval-2024, found that leading open-source detectors lost roughly half their performance compared with older academic benchmarks: 50% for video, 48% for audio, and 45% for images. People don’t do much better. As TRM has written previously, humans identify deepfakes only slightly better than chance.

That’s why detection works best as one layer in a broader response. The US Financial Crimes Enforcement Network (FinCEN) issued an alert in November 2024 warning financial institutions that criminals were using deepfake media to get fraudulent identity documents past verification controls. The alert’s red flags centered on inconsistencies and behavior, not only on the image itself.

Investigators have a structural advantage here. A scammer can fake a face, voice, and identity document. It is much harder to fake the rest of the operation: the payment addresses that receive victims’ money, infrastructure reused across campaigns, and accounts where proceeds are cashed out. When detection fails at the front door, those trails often still lead to the people behind the scam.

What makes law enforcement AI defensible

AI that helps an investigator find a lead is only useful if the evidence behind that lead holds up under prosecution. This is where public sector AI differs most from consumer AI, and where agencies are right to be demanding.

Defensible law enforcement AI has four key traits:

• Provenance: Every finding traces back to its source data, so an investigator can show where a conclusion came from
• A repeatable method: Another analyst running the same steps reaches the same result, in line with established analytic standards such as those used by the US intelligence community
• A human in the loop: The AI surfaces connections; a trained investigator validates them and decides what to act on
• Transparency about limits: Confidence levels are stated, and tools are tested for accuracy and bias in the conditions where they’re actually used

Frameworks to support this already exist. INTERPOL and the UN Interregional Crime and Justice Research Institute (UNICRI) publish a Responsible AI Toolkit for law enforcement, and the UK’s PoliceAI center plans to publish a public register of the AI tools police use. The agencies moving fastest are also the ones building these guardrails in from the start.

‍

{{horizontal-line}}

Frequently asked questions

1. How can law enforcement use AI to fight AI-enabled crime?

Law enforcement uses AI to triage large volumes of digital evidence, resolve identities across data sources, link related cases, trace illicit funds, detect synthetic media, and disrupt criminal infrastructure. The biggest gains come from connecting sources, because AI-enabled crime usually spans identity, cyber infrastructure, and financial activity at once. Effective use keeps a trained investigator in control and documents every step so evidence holds up in court.

2. What AI tools do police use to investigate crimes?

Police use AI for digital evidence review (transcription, translation, and summarization), report drafting, facial and license plate recognition, open-source and social media intelligence, blockchain intelligence, and AI investigation platforms that collate intelligence across many data sources at once. Adoption and policies vary by agency.

3. Can AI detect deepfake scams?

AI can detect many deepfakes, but accuracy drops sharply on real-world content. One 2025 study found leading open-source detectors lost about half their performance on deepfakes circulating online. Detection works best alongside verification procedures and investigation of what scammers can’t easily fake, such as payment trails and reused infrastructure.

4. What is AI-enabled crime?

AI-enabled crime is criminal activity in which artificial intelligence is used to plan, carry out, scale, or conceal an offense. Common examples include deepfake impersonation scams, AI-generated phishing, synthetic identities used to open accounts, and criminal services sold with AI features built in. Scams and fraud are currently the most mature category.

5. Is AI evidence admissible in court?

AI itself isn’t evidence; what matters is whether the underlying findings can be sourced, reproduced, and explained. Courts and prosecutors expect a documented chain of custody, clear method, and qualified person who can testify to how conclusions were reached. That’s why defensible law enforcement AI records provenance at every step and keeps a human investigator responsible for the final judgment.

6. Will AI replace police investigators?

No. AI takes on the repetitive, high-volume work, such as reviewing footage, reconciling records, and running searches across systems, so investigators can focus on judgment, interviews, and building cases. Guidance from the DOJ, INTERPOL, and the UK’s PoliceAI center all emphasizes human oversight of AI in policing.