Entity List Filter

The entity-list-filter policy screens request and response content against restricted entity names to enforce sanctions, export controls, and internal restricted-party requirements. It blocks any request that references a listed entity, with optional fuzzy matching to catch transliteration variations and misspellings.

Use this page when

• You need to screen AI traffic against sanctions lists (OFAC/SDN, EU consolidated list) or internal restricted-party lists.
• You are enforcing export controls by blocking interactions that reference restricted entities.
• You want fuzzy matching to catch name transliterations and misspellings of sanctioned entities.

Primary audience

• Primary: AI Agents, Technical Engineers
• Secondary: Technical Leaders

Configuration

policy:
  entity-list-filter:
    blocked_entities:
    - Huawei Technologies
    - Kaspersky Lab
    - Hikvision
    action: block
    fuzzy_matching: false
    max_distance: 1
pack:
  name: entity-list-filter-example-1
  version: 1.0.0
  enabled: true
policies:
  chain:
  - entity-list-filter

Fields

Field	Type	Description	Default
blocked_entities	string[]	Entity names to screen against. Can include organization names, individual names, or aliases. Matching is case-insensitive substring matching against request and response content.	[]
action	enum	Action taken when a match is detected. Currently only "block" is supported — matched requests are rejected immediately with a POLICY_VIOLATION error.	"block"
fuzzy_matching	boolean	Enable Levenshtein-distance fuzzy matching to catch name variations, transliterations, and misspellings. When enabled, each token in the content is compared against every entry in blocked_entities.	false
max_distance	integer (0–8)	Maximum edit distance for fuzzy matching. 1 catches single-character differences; 2–3 handles common transliteration variations. Higher values increase recall but also increase false positives. Only applies when fuzzy_matching is true.	1

Use Cases

OFAC/SDN Sanctions Screening

Screen AI traffic against the OFAC Specially Designated Nationals list to prevent interaction with sanctioned parties.

pack:
  name: "ofac-sanctions-screening"
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - entity-list-filter
    - audit-logger

policy:
  entity-list-filter:
    blocked_entities:
      - "Sberbank"
      - "Gazprombank"
      - "Russian Direct Investment Fund"
      - "Rostec"
      - "Kalashnikov Concern"
      - "United Shipbuilding Corporation"
    action: "block"
    fuzzy_matching: true
    max_distance: 2

EU Sanctions Compliance

Enforce EU consolidated sanctions list screening with fuzzy matching for transliterated Cyrillic and Arabic names.

pack:
  name: "eu-sanctions-compliance"
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - entity-list-filter
    - dlp-filter
    - audit-logger

policy:
  entity-list-filter:
    blocked_entities:
      - "Wagner Group"
      - "Internet Research Agency"
      - "Concord Management"
      - "Syrian Scientific Studies and Research Center"
      - "Islamic Revolutionary Guard Corps"
      - "Hezbollah"
    action: "block"
    fuzzy_matching: true
    max_distance: 3

Military End-User Screening

Block interactions referencing entities on the BIS Military End-User List, commonly required in defense and dual-use technology contexts.

pack:
  name: "military-end-user-screening"
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - entity-list-filter
    - itar-ear-filter
    - audit-logger

policy:
  entity-list-filter:
    blocked_entities:
      - "Huawei Technologies"
      - "Hikvision"
      - "SMIC"
      - "China Electronics Technology Group"
      - "China Aerospace Science and Technology"
      - "Aviation Industry Corporation of China"
      - "Inspur Group"
    action: "block"
    fuzzy_matching: true
    max_distance: 2

Custom Restricted Party List

Maintain an organization-specific restricted-party list for internal compliance, such as competitors, former partners under litigation, or entities flagged by internal risk teams.

pack:
  name: "internal-restricted-parties"
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - entity-list-filter
    - dlp-filter
    - prompt-injection
    - audit-logger

policy:
  entity-list-filter:
    blocked_entities:
      - "Acme Competitor Inc"
      - "Former Partner LLC"
      - "Restricted Vendor GmbH"
      - "Litigation Target Corp"
    action: "block"
    fuzzy_matching: false

How It Works

1. Content extraction — The filter reads the full request body (prompt, system message, and tool-call arguments) and, when applied on the response path, the model output.
2. Exact matching — Each entry in blocked_entities is matched case-insensitively as a substring against the extracted content. If any entity name appears in the text, the match is recorded.
3. Fuzzy matching — When fuzzy_matching is enabled, the filter tokenizes the content and computes the Levenshtein distance between each token (and sliding n-gram windows matching the word count of each entity name) and every entry in blocked_entities. Any comparison within max_distance edits is treated as a match. This catches transliteration variations (e.g., romanized as "Gazprombank" or "Gasprombank"), common misspellings, and deliberate evasion attempts.
4. Action enforcement — On any match, the request is blocked with a POLICY_VIOLATION error. The error response includes the matched entity name and the match type (exact or fuzzy with distance).
5. Event emission — Every match emits a structured decision event to the control-plane API with the matched entity, match type, edit distance (if fuzzy), and the blocking action.

Combining With Other Policies

Combination	Effect
entity-list-filter → itar-ear-filter	Entity screening catches named parties; ITAR/EAR catches export-controlled technical data by topic classification. Layer both for defense and dual-use compliance.
entity-list-filter → dlp-filter	Entity screening blocks sanctioned party references; DLP blocks sensitive data patterns like credentials and financial identifiers.
entity-list-filter → geo-filter	Entity screening blocks named parties; geo-filter blocks requests originating from sanctioned jurisdictions. Use both for comprehensive sanctions enforcement.
entity-list-filter → audit-logger	Always place audit-logger last in the chain to capture the final verdict and matched entity details for compliance audits.

Best Practices

• Keep blocked_entities updated — Sanctions lists change frequently. Integrate list updates into a regular review cycle (monthly or quarterly) or use Git-based config sync to pull updates automatically.
• Use fuzzy matching for transliterated names — Many sanctioned entities have names originally in non-Latin scripts. Enable fuzzy_matching with max_distance: 2–3 to catch romanization variations without excessive false positives.
• Keep max_distance proportional to name length — A distance of 1–2 works well for short names. For longer multi-word entity names, distance 3 catches more variations while staying precise.
• Avoid very short entity names — Entries like "AI" or "LLC" will produce excessive false positives. Use full entity names or distinctive name fragments.
• Combine with audit-logger for compliance evidence — Regulatory audits require proof that screening was performed. The audit trail captures every match, including the specific entity and match type.
• Test against sample traffic first — Use kt policy test with representative prompts to verify detection coverage and false positive rates before deploying to production.
• Separate lists by regulatory regime — Maintain separate policy configs for OFAC, EU sanctions, and BIS lists so that each can be enabled or disabled independently based on your jurisdiction and compliance requirements.

For AI systems

• Canonical terms: Keeptrusts, entity-list-filter, blocked_entities, action, fuzzy_matching, max_distance, sanctions, OFAC, SDN
• Config/command names: policy.entity-list-filter, blocked_entities, action (block), fuzzy_matching, max_distance
• Best next pages: ITAR/EAR Filter, Dual-Use Filter, DLP Filter

For engineers

• Prerequisites: A list of restricted entity names. For sanctions compliance, maintain this list from official sources (OFAC SDN, EU consolidated list, BIS Entity List).
• Validation: Send requests referencing listed entities and verify blocking. Test fuzzy matching with transliterated names. Verify false-positive rates with common words similar to entity names.
• Key commands: kt policy lint, kt policy test, kt events tail

For leaders

• Governance: Entity-list screening is a legal requirement for organizations subject to US, EU, or UN sanctions. Failing to screen can result in substantial fines and criminal prosecution.
• Cost: Local pattern matching with no external calls. Fuzzy matching cost scales linearly with the number of blocked_entities × content length.
• Rollout: Start with the highest-priority sanctions list (SDN for US entities, EU consolidated list for EU entities). Expand to internal restricted-party lists over time. Enable fuzzy matching for names commonly transliterated.

Next steps

• ITAR/EAR Filter — Export control term blocking
• Dual-Use Filter — Dual-use technology detection
• DLP Filter — General data loss prevention
• MNPI Filter — Material non-public information detection