The FBI's $893 Million Warning: AI Is Now a Primary Cybercrime Vector in Healthcare

COGNOME INSIGHTS | AI GOVERNANCE | THREAT INTELLIGENCE

KEY TAKEAWAYS

• The FBI formally tracked AI-enabled cybercrime for the first time in 2025, attributing over $893 million in losses to AI-assisted schemes with investment fraud, business email compromise, and personal data breaches leading the categories most relevant to healthcare.
• Healthcare and Public Health remains one of the FBI's top three most-targeted critical infrastructure sectors for ransomware, with AI now embedded in attack chains for initial access, lateral movement, and social engineering.
• The Shadow AI visibility gap is the defining governance risk: most AI-enabled crimes go unattributed because victims do not know AI was involved the same dynamic that leaves health systems exposed to ungoverned internal models they cannot monitor or defend.
• The DPRK IT worker scheme poses a direct threat to health systems engaging AI and data science talent, requiring updated vetting protocols for anyone with access to model infrastructure or clinical data.

IN THIS ARTICLE

1. What the FBI's AI Findings Actually Say

2. Healthcare Is the Highest-Value Target

3. Shadow AI: The Threat You Cannot See

4. The DPRK IT Worker Threat: A Direct Healthcare Risk

5. BlackSuit Ransomware and the Healthcare Targeting Pattern

6. The Cost of Inaction: What the FBI Data Implies for Healthcare CFOs

7. What This Means for Your AI Governance Strategy

The FBI's Internet Crime Complaint Center has published its 2025 Annual Report, and the numbers are stark. Americans reported more than one million cybercrime complaints last year, with total losses surpassing $20.8 billion for the first time in the center's 25-year history. That is a 26 percent increase in losses from 2024 alone.

For healthcare CIOs and CISOs, one section of the report demands particular attention: the dedicated chapter on artificial intelligence in cybercrime. For the first time, the FBI formally quantified AI as a distinct threat vector, tracking it across crime types and attributing over $893 million in reported losses to AI-enabled schemes in a single year. The report is unambiguous: AI is no longer a theoretical risk multiplier. It is an operational weapon, actively deployed against enterprises today.

This post unpacks the findings most relevant to healthcare organizations and what they mean for how you govern AI inside your environment.

What the FBI's AI Findings Actually Say

The IC3 report introduced a new 'AI Related' descriptor in 2025, applied to complaints where the victim specifically referenced artificial intelligence as part of the crime. This is likely a significant undercount, many victims of AI-enabled fraud do not know AI was involved yet even on this conservative basis the numbers are material.

AI-assisted investment fraud led all categories, with losses in complaints citing an AI connection surpassing $632 million. Business email compromise schemes using AI-generated voice cloning and text generation accounted for over $30 million in reported losses. Tech support fraud with an AI nexus added another $19 million. Confidence and romance scams using AI-generated personas and scripts generated more than $19 million in losses.

Healthcare Is the Highest-Value Target

Healthcare organizations do not appear as a separate line item in the IC3 report's AI section, but the threat landscape maps directly onto the industry's vulnerabilities. The report identifies Healthcare and Public Health as one of the top three critical infrastructure sectors most impacted by ransomware alongside Critical Manufacturing and Government Facilities.

Healthcare is also structurally exposed to the AI-specific threat vectors the FBI describes. Clinical environments run dozens or hundreds of AI and machine learning models across imaging, diagnostics, medication management, and administrative workflows. Many of those models were deployed by third-party vendors who retain ongoing access. Many were not formally approved through a governance process. And unlike a compromised laptop, a compromised AI model embedded in clinical decision support does not announce itself; it simply returns subtly wrong outputs, for weeks or months, before anyone notices.

The FBI's data on Business Email Compromise is also directly relevant to healthcare. BEC generated over $3 billion in reported losses in 2025. With AI now capable of generating voice-cloned audio indistinguishable from a real CFO or CEO, the social engineering component of these attacks has become dramatically more scalable and convincing.

Shadow AI: The Threat You Cannot See

The most operationally dangerous finding in the IC3 report for healthcare leaders is not a specific crime type. It is the scale of what remains invisible. The FBI's AI-related complaint count of 22,364 cases almost certainly captures a fraction of actual AI-enabled incidents, because victims frequently do not recognize that AI was used against them.

The same visibility problem exists inside your health system. AI models are running across your network right now in vendor software, in tools deployed by clinical teams without IT review, in employee use of public AI platforms that your governance team has not catalogued, assessed, or approved. You cannot govern what you cannot see.

SHADOW AI GROWTH TREND IN HEALTHCARE

2023 25%

2024 55%

2026 85% (Forecast)

An estimated 85% of AI usage in healthcare will be unsanctioned by 2026.

The North Korean (DPRK) IT Worker Threat: A Direct Healthcare Risk

One finding in the 2025 IC3 report deserves specific attention from healthcare CISOs: the North Korean IT worker scheme. The FBI identified dozens of victim organizations where the Democratic People's Republic of Korea dispatched individuals who were hired as remote IT contractors by U.S. companies. These workers used their legitimate access to exfiltrate proprietary data and facilitate further criminal activity.

Healthcare organizations, which regularly engage contractors for EHR implementation, data analytics, AI model development, and interoperability work, are directly exposed to this vector. An AI or data science contractor with access to model training pipelines, clinical datasets, or inference infrastructure represents a high-value target for this type of operation.

BlackSuit Ransomware and the Healthcare Targeting Pattern

The IC3 report documented coordinated disruption actions against the BlackSuit (Royal) ransomware group in August 2025. The group specifically targeted healthcare and public health infrastructure. Modern ransomware operations increasingly use AI-enabled tools for initial access: spearphishing emails generated by large language models, AI-assisted vulnerability scanning, and automated lateral movement once inside a network.

A health system that has deployed AI models without a governance layer has, in effect, created additional attack surface each model endpoint is a potential entry point, and ungoverned models are less likely to have the monitoring instrumentation that would detect anomalous behavior.

The Cost of Inaction: What the FBI Data Implies for Healthcare CFOs

The average loss per IC3 complaint in 2025 was $20,699. For healthcare organizations, total BEC losses across more than 24,000 complaints exceeded $3 billion. The FBI's Financial Fraud Kill Chain team froze over $679 million in fraudulent transfers in 2025 but that represents only a 58 percent success rate, and only for cases where victims reported quickly enough.

These numbers sit alongside the separately documented average U.S. healthcare data breach cost of $10.22 million. Shadow AI incidents add an estimated $670,000 in additional cost per breach event. A health system running 50 ungoverned AI models is not running 50 isolated risks. It is running a compounding exposure that grows with each new deployment.

What This Means for Your AI Governance Strategy

The 2025 IC3 report is not primarily a healthcare document. But read through the lens of a health system CIO or CISO, it is a detailed brief on why AI governance is a security imperative, not just a compliance checkbox.

The report points to several concrete action areas. Discovery must precede governance. You cannot protect models you do not know exist. AI-assisted BEC and voice cloning attacks require updated authentication protocols for financial transactions. The DPRK IT worker findings require rethinking how you vet contractors with access to AI infrastructure. And ransomware's continued targeting of healthcare demands that every model endpoint be treated as part of the attack surface.

The FBI's data quantifies what Cognome has observed directly in production health system environments: the gap between the AI that clinical and operational teams are deploying and the AI that IT and security teams know about is wide, growing, and consequential. Closing that gap is what AI governance operationalized across the full lifecycle is built to do.

Source: FBI Internet Crime Complaint Center, 2025 IC3 Annual Report. ic3.gov. Healthcare breach cost and Shadow AI statistics drawn from industry research cited in Cognome's published materials.

© 2026 Cognome Inc. | cognome.com