Listen to the article
Iran and North Korea are using artificial intelligence to avoid sanctions allowing hostile states to run complex financing schemes with little human involvement, according to a new report by the Royal United Services Institute.
The report called, Algorithms of Evasion: The Rise of AI-Enabled Proliferation Financing, says countries under sanction could use AI to help move money through shell companies, fake documents, cryptocurrency transfers and hidden third-country intermediaries, making it harder for banks and governments to detect illicit payments.
The report warns that AI could “radically increase the scale” of proliferation financing and sanctions evasion, threatening to overwhelm existing detection and enforcement systems.
The report, written by Dr Aaron Arnold, a senior associate fellow at RUSI, says generative AI can produce convincing false documents including driving licences, bank statements and vessel registrations.
The fake documents could then be used to open bank accounts, register front companies or disguise directors and shareholders linked to sanctioned entities.
Generative AI is a type of artificial intelligence that can create new material, such as text, images, audio, video or computer code, after being trained on large amounts of existing data.
RUSI says AI is not necessarily creating wholly new methods of proliferation financing, but is making established techniques faster and more scalable.
Hostile states have long relied on shell companies, third-country intermediaries, fake paperwork and circuitous payment routes to avoid sanctions.
The report says that AI can automate those processes and make them harder to detect.
North Korea is cited as an example of a state already adapting. The report says North Korean IT workers have used generative AI to create online personas, cover letters and CVs, and have used live deepfakes to obscure their identities during online job interviews.
The paper also warns that AI could manage networks of shell companies by generating unrelated names, addresses and contact details, maintaining minimal digital footprints and carrying out small transactions to create “an air of authenticity”.
These networks could remain dormant until activated by triggers such as a new sanctions designation or a transactional need.
Cryptocurrency laundering is another area of concern, according to RUSI. The report says AI-powered systems can analyse blockchain transactions in real time and adjust mixing strategies by changing transaction sizes, timing, pathways and intermediary wallets. This, the report says, could create “an adaptive adversary” that improves its ability to hide illicit funds over time.
The most serious emerging risk, RUSI says, is agentic AI, which could be used to create autonomous systems made up of multiple AI agents, each carrying out separate tasks. Such agents could be given broad objectives, such as moving funds from a sanctioned entity to an offshore account while minimising detection risk, and could then manage shell-company finances, cryptocurrency transfers, deepfake content and social engineering “with little or no human interaction”.
That would challenge traditional enforcement methods built around identifying and disrupting human networks, according to RUSI.
The report says software nodes in an agentic AI network could be “redundant, instantly replaceable and infinitely copyable”, weakening strategies that target key people in illicit organisations.
RUSI says governments should make it easier to use AI to spot sanctions evasion, set clearer rules for cloud and computing providers, and allow banks to share anonymised transaction data to help detect suspicious activity.
The report also says unusual international money transfers should be automatically paused so a person can check them. It says global regulators should update their rules to reflect the risks posed by autonomous AI, while banks and other companies should improve identity checks, watch for deepfakes and use AI to spot suspicious trade documents.
“Ultimately, the shift towards AI-enabled sanctions evasion requires a corresponding shift within public and private sectors, “ the report concludes.
