AI and the algorithms of evasion

Artificial intelligence is transforming the global economy, but the same technology that makes it valuable for businesses may also provide sanctioned states and illicit financial networks with powerful new tools to conceal transactions, obscure ownership, and circumvent international controls. In Algorithms of Evasion, published in May 2026, Aaron Arnold, a Senior Associate Fellow with the Royal United Services Institute's (RUSI) Centre for Financial Crime and Security Studies, examines how AI could accelerate sanctions evasion and proliferation financing activities by states such as North Korea and Iran.

Speaking to The Beiruter, Arnold argued that AI's greatest significance lies in dramatically increasing the speed and scale at which existing techniques can be deployed.

The single biggest capability AI gives actors like North Korea and Iran is the ability to radically increase the speed and scale of sanctions evasion activities to levels that threaten to overwhelm current detection and enforcement capabilities.

 The warning comes as both AI capabilities and sanctions-evasion networks continue to expand. According to Arnold's report, North Korea's Lazarus Group carried out the largest cryptocurrency theft on record in March 2025, stealing more than $1.5 billion from Bybit, while a recent UN Trade and Development report projects that the global AI market will grow from $189 billion in 2023 to $4.8 trillion by 2033. Against this backdrop, Arnold argues that AI could be applied across nearly every stage of sanctions evasion, from generating fraudulent documents and managing shell company networks to laundering cryptocurrency and probing financial systems for vulnerabilities.

Fraud at scale

One of the report's central findings is that generative AI has the potential to overwhelm compliance systems by dramatically increasing the volume of fraudulent documentation that can be produced.

According to Arnold, generative AI can already produce convincing driver's licenses, bank statements, vessel registrations, invoices, and other documents that once required significant time and expertise to create. The challenge is not only their realism, but the speed and volume at which they can be produced.

Early examples are already emerging. The report points to North Korea's overseas IT worker schemes, in which operatives have used generative AI to create online personas, draft resumes and cover letters, and conceal their identities during remote job interviews.

The broader concern is what happens when similar tools are deployed across sanctions evasion networks. A Deloitte study cited in the report estimates that generative AI could enable fraud losses in the United States to rise from $12.3 billion in 2023 to $40 billion by 2027, representing a compound annual growth rate of 32%.

For compliance teams, the problem is fundamentally one of scale.

"Current compliance systems do not have a realistic way of keeping up with that volume," Arnold said.

The use of GenAI to mass-produce high-quality fraudulent documents fundamentally challenges the reliance on document-based verification processes and threatens to overwhelm traditional investigative methods, which are largely manual.

The automation of concealment

Beyond document generation, the report argues that AI could automate one of the most labor-intensive aspects of sanctions evasion, the creation and management of shell company networks.

Historically, concealing ownership required networks of intermediaries, offshore companies, and nominee directors. AI could dramatically reduce that burden by helping build intricate networks of shell companies and fictitious individuals designed to obscure who ultimately controls assets and accounts. Rather than manually constructing these networks, operators could rely on AI systems to optimize corporate structures, identify regulatory gaps, and adapt ownership arrangements in response to enforcement actions.

The report's findings echo concerns raised in a June 2025 report by the Financial Action Task Force (FATF), the global watchdog for money laundering and illicit finance, which warned that sanctions evasion and proliferation financing networks are becoming increasingly complex and difficult to trace.

For Arnold, AI's significance lies less in introducing new concealment methods than in allowing existing ones to operate faster and at greater scale.

Cryptocurrency and adaptive evasion

The report also identifies cryptocurrency as a domain where AI could create new challenges for enforcement authorities. According to RUSI, future AI systems may be capable of analyzing blockchain activity in real time and dynamically adjusting laundering strategies to avoid detection.

For Arnold, this illustrates a broader shift in how sanctions evasion should be understood. Rather than a collection of isolated tactics, he argues that sanctions evasion functions as an adaptive system in which actors continually adjust their behavior in response to enforcement efforts. AI could accelerate that cycle by enabling actors to identify vulnerabilities and react more quickly than human operators.

The result is an ongoing technological competition between enforcers and evaders.

From AI-Assisted to AI-Enabled

While many of the activities described in the report already exist in limited forms, Arnold argues that the most significant risks remain ahead.

The report describes how adversarial AI systems could be used to systematically probe screening systems, identify weaknesses, and develop more effective ways of avoiding detection. In doing so, they could shift sanctions evasion from a reactive process to one that continuously learns and adapts.

For Arnold, however, the greater concern is what happens when those systems no longer require meaningful human direction.

“The most serious AI-related risk over the next five years is the emergence of autonomous AI systems, or agentic AI networks,” Arnold said. 

These autonomous agents could manage complex evasion schemes without direct human intervention, rendering traditional disruption strategies that target human operatives increasingly obsolete.

Recognizing those risks, Arnold argues that enforcement systems must evolve alongside the technologies they monitor. His report calls for enhanced AI-driven screening tools, updated know-your-customer procedures capable of accounting for deepfakes, adversarial testing requirements, and new regulatory standards for autonomous AI agents operating within financial systems.

Whether such measures can keep pace with technological change remains uncertain. What is clear from RUSI's analysis is that the future of sanctions enforcement may depend as much on algorithms as investigators, as regulators confront systems capable of operating at a speed and scale beyond human oversight.