⬇️ Download Tilores Studio for FREE. Run entity resolution locally on your machine.Download free

← Back to Blog
AML / KYC November 9, 2026 · 9 min read

Clean Data on Both Sides: Sanctions Screening With OpenSanctions Founder Friedrich Lindenberg

Steven Renwick
Steven Renwick
CEO, Tilores
Clean Data on Both Sides: Sanctions Screening With OpenSanctions Founder Friedrich Lindenberg

TL;DR: Sanctions screening compares two datasets, and both are messy. Government lists name the same people dozens of times in inconsistent formats, and your own systems hold the same customer several times. In our LinkedIn Live, OpenSanctions founder Friedrich Lindenberg explained how his team cleans the list side. I then showed how Tilores cleans the customer side, screening an OpenSanctions export of about 73,000 rows against a demo bank’s customers in under ten seconds on a laptop.

Book a Demo · Try Tilores Studio (Free)

On this page

  1. Why did Friedrich start OpenSanctions?
  2. What is a sanctions list, and what can be sanctioned?
  3. How is a PEP list different from a sanctions list?
  4. Why is sanctions data itself so messy?
  5. How does an open data project make money?
  6. What did the demo show?
  7. What still can’t be resolved automatically?
  8. Can screening run on-premise or air-gapped?
  9. Do sanctioned people push back?
  10. What should compliance teams take away?
  11. Frequently asked questions

On 5 November I hosted a LinkedIn Live with Friedrich Lindenberg, founder and managing director of OpenSanctions. We had just launched OpenSanctions screening in Tilores Studio, and I wanted to hear from the person who has spent years on the other side of the problem: the sanctions and PEP data itself.

Friedrich joined from New York at an early hour, and the conversation covered more ground than I expected, from Chinese sanctions policy to a phone call with a sanctioned car mechanic. Here are the parts most useful to anyone who runs, buys or builds screening. Quotes are lightly edited for readability, and you can watch the full recording on LinkedIn.

Why did Friedrich start OpenSanctions?

OpenSanctions began as a research tool for investigative journalists, not as a compliance product.

Friedrich spent years working with reporters in Central and Eastern Europe on corruption investigations. Again and again, the same small group of people held both political and economic power and used one to grow the other. “If you’re investigating that,” he said, “one thing that’s really useful to have is a baseline record of who the people involved are.” Those people were either political office holders or named on an international sanctions list.

Since around 2022, commercial use has grown fast. Companies in supply chain risk, identity verification, anti-money laundering and transaction monitoring now use the data as a risk signal in their own processes.

I asked why OpenSanctions chose to be a data resource instead of building a full compliance product. His answer comes from open source thinking: “We are building one building block of a compliance stack. That building block is the data problem that we are really competent to solve.” Other companies then combine it with their own components to fit their regulatory or intelligence needs.

What is a sanctions list, and what can be sanctioned?

Friedrich called sanctions lists “a really interesting hack of international law”.

If someone commits serious wrongdoing inside your country, you prosecute them. If they’re abroad, perhaps running another country, you can’t. So a government does the next best thing and makes it a crime for its own industry to deal with them. Banks in that country may no longer provide services, facilitate investments or send goods to anyone on the list. Every regulated firm then has to check whether any customer or counterparty is a listed party or directly linked to one.

People are only part of it. “A boat can be on a sanctions list. An address can be on a sanctions list, as can a crypto wallet,” Friedrich said. That variety means you need a proper taxonomy just to describe what has been sanctioned.

The volume is significant. The EU alone publishes 45 different sanctions lists, and OpenSanctions brings them together into one resource. As a public-interest resource rather than a commercial product, it also tracks lists from countries such as Russia, China and Iran. Friedrich pointed out that China has started running its own sanctions policy aimed at specific parts of the US defence technology sector.

How is a PEP list different from a sanctions list?

A politically exposed person (PEP) isn’t banned from anything. They just call for closer scrutiny.

Friedrich’s example: if a politician walks into your bank with a plastic bag full of cash, that’s a different risk from a private citizen doing the same. Public power raises the odds that money came from bribery, procurement fraud or other corruption. Banks have to check where a PEP’s money comes from and whether the amount is plausible. In the UK, the best-known recent example is Nigel Farage, who was flagged as politically exposed and whose account at Coutts was closed in 2023, which became a public controversy. But being a PEP doesn’t mean a bank refuses you. Otherwise no member of parliament could have a bank account. In some countries refusing on that basis alone is illegal; banks are instead expected to apply a lower threshold for flagging suspicious activity.

PEP data is also much harder to collect. Sanctions lists are published by governments. According to Friedrich, only six or seven governments publish PEP data in any coherent form. The rest is “a collector’s effort”: OpenSanctions models which positions make someone politically exposed, from head of government to head of a state electricity company or deputy governor of a central bank, and then tracks who holds each position at any given time.

That job is about to get bigger. When the EU’s new anti-money laundering regulation applies from July 2027, PEP coverage has to go much deeper. Friedrich’s example was that the mayor of every city with more than 50,000 inhabitants will count as politically exposed. We cover what that means for bank customer data in our EU AMLR readiness guide.

Newsletter
Enjoying this article? Get the next one in your inbox.

New posts on entity resolution, identity data and AI. No spam, unsubscribe any time.

By subscribing you agree to our privacy policy.

Why is sanctions data itself so messy?

Government data is inconsistent, and the same people appear on many lists.

OpenSanctions maps every source into a structured data model called Follow the Money. It covers names, addresses, dates of birth, tax identifiers and everything else that helps identify a person, company or vessel. Along the way it fixes problems in the source data. A common one, Friedrich said, is a government putting several names for the same person in a single name field, which then has to be split.

The bigger problem is overlap. Western allies’ sanctions lists contain largely the same individuals, companies and vessels. Without deduplication, a customer who resembles one sanctioned person could trigger “34 different alerts” instead of “one alert that says this is a risk that might be linked to 34 different sanctions lists”. That’s entity resolution, the same problem Tilores solves for customer data.

Friedrich was modest about the scale on his side: there are roughly 100,000 sanctioned entities worldwide, “so at some point you start knowing them by name”. Customer databases are far larger, noisier and usually hold less information per record.

The data also moves quickly. Somewhere in the world a sanctions list changes every couple of hours. OpenSanctions has seen about 700 updates to US sanctions lists alone since it started. It pulls each source every few hours and publishes a fully integrated snapshot of the whole database every six hours.

How does an open data project make money?

By selling reliability, not access.

Anyone can search OpenSanctions on its website or download the entire dataset as a single file. “We want to be easy to play with,” Friedrich said. Once a business depends on the data, it needs a supplier relationship: a guarantee that the data will be updated tomorrow, support, and someone to explain the material. That’s the commercial offer. Companies can buy screening volume through the OpenSanctions API, and technology companies building AML, KYC or supply chain risk platforms license the bulk data. OpenSanctions now combines around 500 sources into one dataset.

What did the demo show?

The second half of the session was a live demo in Tilores Studio, our desktop application, running in local mode on my laptop. The data was a demo bank with three simulated internal systems.

Step 1: resolve the customer data. Two systems were already loaded. I ingested the third, the core banking system, as a CSV. Studio mapped most of the columns automatically and linked the new records to the existing ones at about 1,200 records per second. That’s one process on one laptop; in the cloud, ingestion runs in parallel at thousands of records per second.

Every customer becomes an entity with a golden record, the best available data across all its source records. You can open the entity as a graph to see every record, which system it came from and why two records were linked, including transitive links where a third record connects them. A quick search for Nigel Farage, who also appears in the demo data, returned one entity built from three records in different systems.

Step 2: screen against OpenSanctions. I then loaded an OpenSanctions export of about 73,000 rows, covering people, organisations and at least one vessel. Studio recognised it as OpenSanctions data, applied the recommended mapping and screened it against every customer entity at about 7,000 rows per second. The full screening finished in under ten seconds. 364 rows matched, hitting 495 customer entities.

Two results showed what resolving first is good for:

  • Vladimir Putin matched an entity made of three records from different systems, and the sanctions record hit all three. The match came from name similarity, and Studio also showed the attributes that agreed exactly: country, last name and date of birth.
  • Alexander Lukashenko matched two separate entities that were clearly the same person but had not been merged under the demo’s configuration. With the compare view I could check them side by side and send the pair to the review queue, where a person or an AI agent confirms the merge.

In production all of this runs through the API. You can feed in OpenSanctions deltas or the full daily dataset, keep the resolved customer data continuously up to date, and push results into your own systems. Studio then serves as the screening and data stewardship interface.

Friedrich’s reaction went straight to the edge cases: “Showing that comparison view of the two Lukashenko records is so helpful, in terms of taking the obvious cases, but then also taking the care and diligence to really make sure that the edge cases are being handled just the way you want them.” On the sanctions side, he said, those edge cases include names spelled in different alphabets and lists where someone has swapped the day and month of a date of birth.

What still can’t be resolved automatically?

People with the same name and too little data.

Friedrich’s example was Donald Trump and his son. The media calls the son Donald Trump Jr., but that’s not what his passport says. To tell them apart you need other attributes, such as an address or a date of birth. “If you have enough information, everything is possible. If it gets sparse, then that’s a real problem.”

He also admitted that OpenSanctions has the same problem with its own customers. More than 10,000 people have signed up to try its API, and their details are now spread across the CRM and several other systems with no central view. That’s the pattern we see everywhere: the same person held in several systems, and often several times within one system.

Can screening run on-premise or air-gapped?

Yes. A viewer asked this in the comments, specifically about Red Hat Enterprise Linux.

Tilores is AWS-native by default and infrastructure-agnostic by design. It runs on Azure, Google Cloud, a single server or fully air-gapped on-premise, including on RHEL. The advantage of AWS serverless is that scaling up and down is handled for you. On-premise, your team takes on databases, backups and stability. Read more on our deployment options page.

Friedrich added that OpenSanctions bulk data fits this model well. Because customers download the full dataset, screening can happen entirely inside their own environment. “You never have to tell us who your customers are,” he said. “We just provide you with the feed of the baddies.”

Do sanctioned people push back?

Often, and Friedrich has learned to treat it “almost like a QA department”.

OpenSanctions receives a steady stream of complaints and legal pressure from people unhappy to appear on its site. About half of those turn into useful discussions about whether OpenSanctions represents the official listing accurately. The result is usually a better data model or clearer presentation, which may not satisfy the person complaining. “It’s a really helpful way for us to have the most critical audience in the world,” he said, “and therefore really up our game on data provenance, on data lineage, and also on data quality.”

He also told the story of an early phone call with a car mechanic who had been placed on the UN sanctions list, for reasonably good reason, and had since served his time. OpenSanctions doesn’t decide who is listed, but Friedrich helped him work out how to approach the UN about having the listing reviewed.

What should compliance teams take away?

  • Screening quality depends on both datasets. Clean, deduplicated sanctions data still produces weak or duplicate alerts if your customer data is fragmented.
  • Resolve first, then screen. One entity per real customer means one alert per real risk, based on every attribute you hold.
  • Plan for frequent change. Sanctions lists change several times a day, and PEP coverage is about to widen under the AMLR.
  • Keep a human, or an agent, for the edge cases. Same names, transliterations and data entry errors need a review queue, not just a threshold.
  • Screening can stay in-house. Bulk sanctions data plus a self-hosted or own-cloud Tilores deployment keeps your customer list inside your environment.

You can try the whole workflow yourself: download Tilores Studio for free and load the OpenSanctions data. Watch the full recording on LinkedIn to see the conversation and the demo end to end.

Frequently asked questions

What is OpenSanctions?

OpenSanctions is an open database of sanctioned entities, politically exposed persons and related risk data. It brings together around 500 sources into one consistent dataset. Anyone can search or download it, and businesses can buy API screening volume or license the bulk data.

How often does sanctions data change?

According to Friedrich, a sanctions list changes somewhere in the world every couple of hours. OpenSanctions pulls each source every few hours and publishes an updated snapshot of the full database every six hours.

What is the difference between a sanctions list and a PEP list?

Sanctions lists are published by governments and make it illegal to do business with the listed parties. PEP lists identify people who hold or held public office, and their close associates, who call for enhanced due diligence but are not banned. PEP data is largely collected by data providers because few governments publish it.

How fast is sanctions screening in Tilores Studio?

In the live demo, Tilores Studio running locally on a laptop screened about 73,000 OpenSanctions rows against a demo bank’s customer entities at about 7,000 rows per second, finishing in under ten seconds. The cloud version runs in parallel and is faster.

Can I screen without sending my customer data to a third party?

Yes. You can download the OpenSanctions bulk data and screen it in Tilores Studio locally, or in a Tilores deployment in your own cloud account or on-premise, including air-gapped environments.

See what resolved entity data does for your business — and your AI.