Sumsub uses LLMs (Large Language Models) in its AML Screening solution as an additional analysis layer during AML match review.

LLM models help:

• Analyze the context of potential AML matches.
• Identify inconsistencies between applicant data and matched record data.
• Generate reasoning notes that explain match classification.
• Provide additional input to Sumsub’s decisioning framework, which determines the final match state.

📘

Note

The AML screening flow and match outcomes remain the same. The main difference for clients is the addition of transparent reasoning notes.

If you keep LLM support enabled, LLM models complement existing rule-based checks by performing deeper contextual analysis across supported AML match types.

Identity and name matching

For identity and name matching, LLM models assess whether the applicant and the matched record likely refer to the same person. To do this, they:

• Compare full names and individual name parts, including first, middle, and last names.
• Evaluate name similarity and spelling distance.
• Consider initials when comparing names.
• Detect missing or additional name parts.
• Estimate gender when relevant.
• Compare full or partial dates of birth.
• Compare country data linked to both the applicant profile and the matched record.

PEP matching

For PEP (Politically Exposed Persons) matches, LLM models verify whether the match is logically consistent. They:

• Review when the person started holding the political position.
• Verify that the person’s age at that time is plausible.
• Reject matches where the timeline indicates the person would have been underage.

Adverse media matching

For adverse media matches, LLM models review the source context and timeline. They:

• Identify whether the person in the article is the subject of the report rather than a victim, witness, or reporter.
• Verify age consistency when the source includes age-related details.
• Check publication dates to confirm that the timeline is consistent.

Fitness, probity, and warning lists matching

For fitness and probity or warning list matches, LLM models apply logical consistency checks. They:

• Verify age and timeline consistency.
• Consider the date when the list included the person.
• Rule out matches that contain logical inconsistencies.

AML Screening LLM models FAQ