Part I. AI’s Crucible Moment

When you think of “crime,” what comes to mind? For most of us, it’s burglars, pickpockets, and car thieves. The biggest and fastest-growing threats, however, are digital.

Between 2020 and 2025, digital crime losses reported to the FBI rose fivefold to USD 20.9 billion, while the estimated number of US property crimes fell nearly 20%.²

Crimefighters have responded aggressively. Working with banks, the FBI’s Financial Fraud Kill Chain froze a record USD 679 million in 2025, up 21% from 2024. Despite often lacking dedicated cybercrime units, many US state and local agencies are also acting: Virginia’s Loudoun County Sheriff’s Office seized USD 1.4 million stolen in a cryptocurrency investment scam in 2025.³

Yet crime is outpacing the response. Gallup estimates US adults lost USD 68 billion to scams in 2025.⁴ We figure that US law enforcement freezes, recovers, or seizes only 2–4% of those losses, annually.⁵

This is the capacity gap: the difference between the crime a society experiences and what law enforcement has the capacity to investigate or prevent.

The capacity gap does not reflect a failure of law enforcement. The problem is structural.

At its core, the capacity gap exists because of the advantage criminals have in speed, cost, and scale. The gap has been growing for years, and AI is accelerating it:

• Speed: 130 seconds vs. 576 days. A study found AI voice agents could carry out common phone scams in 130 seconds; the median US federal fraud case takes 576 days to reach a charging decision.⁶
• Cost: 24 to 1. Our illustrative model estimates a scam compound’s cost at roughly USD 6,000 per victim, compared with roughly USD 145,000 in aggregate costs per fraud charge.⁷
• Scale: 0.45% of recorded frauds charged. Of 1.23 million frauds recorded in England and Wales in the year ending March 2025, about 1 in 24 reached a police force for investigation and 1 in 220 led to a charge.⁸

A foundational problem is that criminals and law enforcement operate under different constraints. Criminals adopt new technology the moment it’s available, work from any country, and answer to no one. In their pursuit of money, they act with abandon and without concern for collateral damage. Law enforcement, on the other hand, must operate within defined jurisdictions, secure funding and approval before adopting new technology, and build cases that can withstand scrutiny in court.

The capacity gap threatens public safety, and hiring or budget increases alone cannot close it. Law enforcement needs the tools to keep pace with criminals while upholding the principles they're duty-bound to protect.

Ardence is the first operating system for crimefighters, and we built it to help close the capacity gap.

Already in use across law enforcement, defense, and private sector organizations, Ardence has helped investigators scale World Cup security, disrupt AI-powered cybercrime services, and take down thousands of scam sites.

Today, the intelligence required to solve a case is scattered across platforms that don’t talk to one another. Investigators search one database, copy results into another tool, and repeat across a dozen tabs. Hours and days go into gathering information and reconstructing what happened. Meanwhile, the money moves, the infrastructure changes, and more people become victims.

Now, on Ardence, investigators can combine intelligence across connected sources, automate time-consuming workflows, and share findings with authorized collaborators in one powerful, AI-powered platform.

Source provenance helps investigators verify findings against underlying records. Investigators set objectives, control consequential decisions, and remain responsible for corroborating AI-generated leads before taking action. Ardence empowers them to move at machine speed while maintaining guardrails and auditability.

Today's investigators know how to build a case. They need the ability to do it at the scale and speed of the threats they face.

Put simply, criminals are using AI to move at the speed of thought. With Ardence, crimefighters can, too.

In naming our new flagship platform, we thought about the people we built it for: law enforcement officers, compliance professionals, intelligence analysts, and cybersecurity experts. For them, protecting victims and fighting crime is more than a job; it’s a calling.

In them, we see total dedication to a mission worth fighting for. They are committed to bringing justice to those who have been wronged and building a safer world for all.

Today’s crimefighters are, in a word, ardent—passionate, energetic, and all in for the cause. That quality inspired the name Ardence.

Ardence gives the world’s defenders more capacity to act on their resolve: to follow more leads, disrupt more criminal networks, and protect more victims.

Today, we honor that resolve.

Esteban Castaño
CEO and Co-Founder

1. 1. AI adoption. TRM Labs, The 2026 AI-in-Crime Adoption Index, August 17, 2026. The 40% figure refers to year-over-year growth in TRM’s composite index of AI adoption across four crypto-enabled crime categories. The index combines prevalence, breadth of use across the crime lifecycle, and sophistication; it is not a count of criminals or crimes.
2. 2. Digital crime and property crime, 2020–2025. FBI, 2020 Internet Crime Report (opens in a new tab), p. 15, and 2025 Internet Crime Report (opens in a new tab), pp. 4–6: reported losses increased from approximately USD 4.2 billion to USD 20.877 billion, or roughly fivefold. FBI 2020 crime statistics (opens in a new tab) and Summary of Reported Crimes in the Nation, 2025 (opens in a new tab), p. 8: estimated US property crimes fell from 6,452,038 to 5,245,768, a decline of 18.7%. Both comparisons cover 2020–2025; one measures reported financial losses, the other estimated offenses. IC3 accepts international complaints as well as US complaints.
3. 3. Law enforcement response. FBI, 2025 Internet Crime Report (opens in a new tab), p. 17: USD 679,013,183 frozen through the Financial Fraud Kill Chain, compared with USD 561.6 million in 2024 (2024 Internet Crime Report (opens in a new tab), p. 13), an increase of 20.9%. Loudoun County Sheriff’s Office, quarterly report presented May 6, 2025 (opens in a new tab), p. 16, documents the approximately USD 1.4 million seizure announced February 24, 2025. Evidence of uneven access to cybercrime personnel comes from the Bureau of Justice Statistics’ Local Police Departments Personnel, 2020 (opens in a new tab), Table 13, and Primary State Law Enforcement Agencies Personnel, 2020 (opens in a new tab), Table 10. These staffing figures describe the sector; Loudoun itself had a Financial and Cybercrime Unit.
4. 4. Estimated US scam losses. Gallup and Stop Scams Alliance, United States of Scams: The Financial and Emotional Fallout (2026), summarized in “Scam Victims Report Billions Lost and Harm to Mental Health” (opens in a new tab), June 29, 2026. The USD 68 billion estimate covers US adults personally scammed in 2025, based on a nationally representative survey of 5,173 adults conducted January 8–February 18, 2026. It is a survey estimate, distinct from losses reported to law enforcement.
5. 5. Scale of freezes and recoveries. TRM’s The Capacity Gap, endnote 2, identifies USD 1.429 billion in freezes and recoveries across two federal programs in 2025, equivalent to 2.1% of Gallup’s estimated USD 68 billion in US scam losses. Its broadest accounting totals approximately USD 6.45 billion over 2023–2025, averaging USD 2.15 billion annually, or 3.2% of the Gallup figure. The illustrative 4% scenario allows USD 2.72 billion annually—approximately USD 569 million, or 26%, above that broad average—for additional enforcement activity outside the accounting. This allowance is an assumption, not an observed total or a measured national upper bound. The broad accounting includes seizure categories with uncertain links to scam losses and possible overlap. Reporting periods and victim populations differ, so these figures compare relative scale rather than establish the share of Gallup’s losses recovered. Funds frozen or seized are not necessarily returned to victims.
6. 6. Speed. Richard Fang, Dylan Bowman, and Daniel Kang, Voice-Enabled AI Agents can Perform Common Scams (opens in a new tab), October 21, 2024, Table 2. Successful scam tasks in a controlled experiment took 77–183 seconds; 130 seconds is the midpoint of that range, calculated for this comparison, rather than the study’s reported average. Researchers simulated cooperative victims; the experiment tested execution rather than real-world persuasion. Bureau of Justice Statistics, Federal Justice Statistics, 2023 (opens in a new tab), Table 4: the median time for federal fraud matters concluded in fiscal 2023 was 576 days from referral to a US attorney’s decision to prosecute or decline. The comparison illustrates different stages: executing a scam task and reaching a prosecutorial decision.
7. 7. Cost. TRM’s The Capacity Gap, endnote 7, models a Southeast Asian scam compound’s operating cost at USD 5,989 per victim defrauded, rounded to USD 6,000. This is a modeled estimate, not an observed industry average. The public-sector figure is calculated from the UK Home Office’s Economic and Social Cost of Fraud 2023 to 2024 (opens in a new tab), Table 5: GBP 179 million in law enforcement costs plus GBP 243 million in criminal justice costs for fraud against individuals. Dividing GBP 422 million by 3,699 charge/summons outcomes in the year ending March 2024 (Crime Outcomes in England and Wales 2024 to 2025 (opens in a new tab), Table 5.2.1) gives GBP 114,085, or approximately USD 145,000 at USD 1.27 per GBP. The resulting ratio is about 24:1. It compares modeled criminal costs with aggregate public response spending per outcome, not the cost of prosecuting an individual case.
8. 8. Scale. UK Home Office, Crime Outcomes in England and Wales 2024 to 2025 (opens in a new tab), Tables 5.1 and 5.2.1. In the year ending March 2025, 1,225,778 fraud offenses were recorded, 50,715 were referred to police, and 5,530 received a charge or summons outcome. Dividing referrals and outcomes by recorded offenses gives approximately 1 in 24 and 1 in 220, respectively; 5,530 ÷ 1,225,778 = 0.45%. These are annual totals, not a tracked cohort: some referrals and outcomes concern offenses recorded in earlier years.