AI in Sanctions and PEP Screening: How ML and NLP Strengthen Compliance

Screening users against official lists of sanctioned parties and Politically Exposed Persons (PEPs) is a key compliance requirement in most jurisdictions. The penalties for getting it wrong can be severe, as British financial services company Markom Management Limited learned when it was fined £300,000 (approx. USD $400,000) in 2025 for breaching UK sanctions against Russia.

Traditionally, checking users against sanctions lists has been a labor-intensive process. Companies need to verify a customer’s identity, then check various watchlists, not just for the name they have given, but also variations on that name, and other names they might have used.

Sanctions screening must be meticulous to avoid any errors that could result in a sanctioned individual being taken on as a customer or a PEP not being properly flagged as high-risk. However, it’s also essential to minimize false positives, where a user is incorrectly flagged as being on a sanctions list or a PEP because they share a similar name to someone who is.

Now, the latest use of AI in compliance is set to bring sanctions screening right up to date. The process will become simpler and more accurate, reducing false positives and cutting workloads, while maintaining robust sanctions compliance.

Let’s explore how AI is being deployed in sanctions and PEP screening to make the process more efficient and effective.

How AI works in screening workflows

Sanctions and PEP database and watchlist screening are a key part of more general AML screening workflows, into which AI is increasingly being integrated.

The following is an example of how AI can be used for compliance automation, specifically in relation to watchlist screening:

Machine Learning, Natural Language Processing, and entity resolution

Deploying machine learning (ML) in compliance processes means the tools we use can become ever better at spotting patterns and making accurate predictions by learning from the data they have previously processed. 

Large Language Models (LLMs), a new generation of NLP models, are the next step in the evolution of AI-powered compliance. Because LLMs are advanced NLP models trained to understand and generate human language, they can analyze broader contextual information for tasks such as name screening, which allows them to more accurately determine whether a match to a watchlist is genuine.

For example, an LLM can consider factors such as whether a user’s verified birth date and country of residence might rule them out as the person they have been matched with on a particular list.

This sort of context-sensitive analysis can be particularly beneficial for ‘entity resolution’, in which different pieces of information are processed to determine whether they can be attributed to the same individual or not.

Reducing false positives at scale

False positives can occur in sanctions screening when a user’s name is incorrectly flagged as matching someone on a watchlist, for example, because they have a similar name and live in the same area.

False positives during sanction screening can waste a lot of time and resources, and could mean legitimate customers are unfairly denied services. Automated sanctions screening can help reduce this by rapidly checking a range of contextual material to determine whether a match is likely to be genuine.

Why false positives are costly

Every match flagged during sanctions screening costs organizations time and money, as it must be reviewed to determine whether it is genuine. For legitimate matches, this is time and money well spent, but when there are many false positives, the cost can become very significant.

False positives can also increase onboarding times for customers and may even lead them to choose to take their business elsewhere, costing a service provider even more.

Following sanctions screening best practices can help reduce false positives by ensuring you have suitable compliance tools. 

Sanctions risk assessments can also be useful to understand your company’s sanctions risk exposure, so you can set screening criteria that meet compliance requirements without being so sensitive that they trigger excessive false positives. 

How AI cuts alert noise

AI compliance tools can significantly cut false positives by checking potential matches using contextual material, such as subjects’ ages and address data, to rule out any obvious mismatches. This can cut the risk of ‘alert noise’ caused by multiple false positives that can tax compliance teams and increase the risk of genuine matches being missed.

A reduction in sanctions false positives can also be achieved by applying flexible screening criteria based on a subject’s risk profile. Higher-risk individuals can undergo enhanced checks, while lower-risk individuals are subject to more streamlined screening, reducing the likelihood of incorrect watchlist matches.

AI-powered name matching

Name screening has to have a degree of flexibility. There can be many legitimate reasons why people might not use the same name in every context, for example, if they sometimes go by a nickname or have changed their name. 

❗However, criminals may also use variations of their name or different names entirely to attempt to circumvent name screening and other AML checks.

Name matching algorithms must use techniques such as ‘fuzzy name matching’ that allow them to recognize that close variations of a name may refer to the same person and should be flagged for further review. This can significantly increase the accuracy of name screening.

Multilingual name challenges

Sanctions list screening and Politically Exposed Person (PEP) screening must account for people’s names being written differently in different languages and scripts. Machine learning compliance tools can assist with this by ensuring that all possible variations of a name are rapidly and accurately flagged and sent for further review when needed.

Phonetic and fuzzy matching

Name-matching algorithms used for name screening should employ phonetic and fuzzy matching for names. This means they can recognize the relationship between names that sound similar but are spelled differently (e.g., ‘John’ and ‘Jon’), as well as common variations of names (such as ‘Paco’ being a nickname for ‘Francisco’).

NLP for contextual risk detection

In financial crime detection, LLMs can review vast volumes of data in seconds. They can quickly identify broader risk signals, for example, by performing multilingual adverse media screening to detect news indicating a customer may pose an elevated risk. This enables faster and more comprehensive analysis of potential risk factors than manual review alone.

Extracting risk signals from text

Adverse media screening relies on the ability to connect subjects with references across different media, then analyze those references to determine whether they indicate a higher risk of involvement in financial crime. 

Adverse media and PEP screening tools can use NLP to scan media sources and flag potentially relevant content, such as references to criminal investigations and convictions, then surface these in a human-readable format. NLP can also be used to analyze adverse media references to differentiate whether a subject was the perpetrator of a particular action, or perhaps a victim, reporter, or otherwise involved.

Real-time monitoring and alerts

Sanctions screening is not a one-and-done activity: customers must be continually re-screened through sanctions monitoring to check whether they have been added to a watchlist or if new information has come to light connecting them to a sanctioned party.

AI-powered tools can enable continuous monitoring of sanctions lists, as well as real-time sanctions screening for new and existing customers.

Continuous compliance benefits

AI-powered real-time sanctions screening offers continuous compliance with regulatory requirements. This helps ensure that any new information related to sanctioned parties is quickly identified and checked against relevant customers, so appropriate action can be taken promptly.

Continuous sanctions compliance reduces the risk of breaches that could result in fines and negative publicity. It should be deployed alongside real-time transaction monitoring to ensure an unbroken chain of compliance with regulatory requirements.

Operational and regulatory benefits

Integrating AI into sanctions screening solutions can offer many of the benefits seen with other compliance automation tools.

Efficiency and cost savings

Compliance automation tools can rapidly process vast amounts of data, completing tasks faster and reducing resource requirements. Compliance automation can result in massive wins, such as the average 70% reduction in case resolution time and $2.6 million savings achieved by clients using Sumsub’s AML compliance software.

Auditability and regulatory alignment

AI-powered tools can support sanctions screening best practices by automating labor-intensive processes essential to audits, such as documentation and recordkeeping, thereby supporting better human decision-making. They can also ensure regulatory alignment in different jurisdictions by allowing models to be adapted to the specific requirements of each jurisdiction, for example, following the Office of Foreign Assets Control (OFAC) screening process that is mandatory in the US.

Global and multilingual coverage

Effective watchlist screening should involve checking subjects against global watchlists and sources that may be in multiple languages. Compliance tools with LLM capabilities can ensure PEP screening and sanctions list screening are not compromised by relying solely on local watchlists and new sources in a single language, without imposing a huge administrative burden on compliance teams.

AI governance and ethics in screening

In 2026, a growing number of countries have developed, or are in the process of developing, AI policies and regulations. Therefore, businesses should prioritize regulatory compliance, too, such as EU AI Act compliance, when implementing AI tools.

Observance of AI governance and ethics is important for compliance purposes, as it ensures you implement AI tools responsibly, benefiting from AI’s benefits while mitigating any risks. Carrying out an AI risk assessment is highly recommended before the deployment of any new AI tool in compliance workflows to help understand any compliance risks that might arise, and what steps you need to take to address them.

EU AI Act and compliance frameworks

The EU AI Act is a trailblazing piece of legislation that aims to ensure AI systems that can affect the lives of EU residents are “safe, transparent, traceable, non-discriminatory and environmentally friendly”. 

Any company using AI systems in compliance workflows that impact people living in the EU will need to understand the EU AI Act compliance obligations and ensure their risk management framework accounts for them.

Supervised vs unsupervised learning

AI compliance tools must produce reliable results that organizations can trust; otherwise, their benefits are outweighed by risks such as a lack of transparency or explainability, and even regulatory breaches. Understanding the difference between supervised and unsupervised learning is therefore essential when applying machine learning for compliance.

Supervised learning involves training models on labeled datasets prepared by compliance specialists. The system learns how specific inputs should be classified—for example, distinguishing legitimate transactions from suspicious ones—and can then apply those learned patterns to new data.

Unsupervised learning, by contrast, analyzes unlabeled data to identify hidden structures or anomalies on its own. This approach is commonly used to detect unusual behavior patterns that may not match known typologies of financial crime.

In practice, effective compliance platforms combine both approaches: supervised models provide consistent decision-making aligned with regulatory expectations, while unsupervised models help uncover previously unknown risks. 

❗However, because AI systems can raise transparency and explainability challenges, organizations must ensure strong accountability over AI-driven workflows, including mechanisms for human-in-the-loop and human-on-the-loop oversight.

Preparing for AI-enhanced screening

As with all AML compliance software, proper preparation is needed before implementing new sanctions screening solutions that are enhanced with AI.

Integrating AI with AML and KYC

Sanctions screening is a part of the wider process of Know Your Customer (KYC) in AML compliance. KYC automation can allow various other parts of the process to be automated, such as ID verification. 

Integrating AI into KYC tools more generally can help to ensure that individual parts of the process, such as AML screening, can be carried out quickly and efficiently, with all results seamlessly aggregated for review and recordkeeping.

Training compliance teams on AI

Compliance teams will need specialist training on AI compliance tools to ensure they understand their capabilities and how human intervention fits into newly automated processes. Continuous team training should be part of your sanctions screening best practices to make sure everyone stays abreast of the latest requirements and can get the best out of the tools at their disposal.

Moreover, the EU AI Act (Art. 4) introduces an “AI literacy” requirement, meaning organizations must ensure staff have sufficient knowledge to use AI systems responsibly.

Suggested read: Comprehensive Guide to AI Laws and Regulations Worldwide (2026)

How Sumsub can help

Following the successful integration of AI with many of our solutions, Sumsub is now adding LLM support to our AML screening tool. This will streamline AML alerts and case reviews, saving analysts and compliance teams valuable resources and time.

LLM support for our AML screening solution will enable deeper contextual analysis on potential watchlist matches, improving accuracy and reducing false positives. LLMs' outputs will be made available in a human-readable note format for easy manual review, which allows informed human decision-making.

While our LLM-powered AML screening can save time and effort, it never cuts corners. If there is not enough information to make a firm judgment on whether the identity you are checking matches a watchlist, it will always say so. It is also particularly careful with sanctions lists, given the severe potential consequences of any errors. This means you can have full confidence that you can reap the benefits of screening automation while staying compliant.

Identify AML risks with Sumsub

Detect high-risk users by screening them against global watchlists for sanctions, PEPs, and adverse media.

Try our AML Screening now

FAQ: AI in sanctions and PEP screening

• What is sanctions screening?

  Sanctions screening is the process companies must go through to check their users against official lists of sanctioned individuals, companies, and other parties issued by governments and international organizations.

• What is PEP screening?

  PEP screening is where companies check their customers against lists of Politically Exposed Persons (PEPs) such as politicians, judicial and military officials, and executives of state-owned corporations. These people are considered to be at higher risk of bribery, corruption, and money laundering, so this must be considered as part of their risk profile.

• What is AI in sanctions screening?

  AI for sanctions screening refers to the use of AI technology, such as Large Language Models (LLMs), to make checking users against sanctions lists more efficient and effective. These tools can rapidly check multiple lists and name variations using contextual information to improve name matching and support more informed human decision-making.

• How can AI reduce false positives?

  AI can reduce false positives in sanctions and PEP screening by using contextual information to determine whether a user genuinely matches someone on a watchlist or simply shares a similar name. This contextual information can include things such as gender and dates of birth.

• Do regulators allow AI in sanctions screening?

  AI in sanctions screening is generally allowed by regulators. However, some jurisdictions are starting to introduce legislation, such as the EU AI Act, that sets compliance requirements for the use of AI applications, meaning that if the AI tools deployed in section screening are regulated by the EU AI Act, you should take compliance requirements into account.