Dual-Use Filter

The dual-use-filter policy detects references to dual-use items — goods and technologies that have both civilian and military applications — to enforce export control compliance.

Use this page when

• You need to detect references to dual-use items that have both civilian and military applications.
• You are enforcing export control compliance under the Wassenaar Arrangement or EU Dual-Use Regulation.
• You want to combine dual-use detection with ITAR/EAR and entity-list screening in a defense or research environment.

Primary audience

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

Configuration

policy:
  dual-use-filter:
    blocked_terms:
    - centrifuge
    - precursor chemical
    - maraging steel
    - detonator
    action: block
    fuzzy_matching: false
    max_distance: 1
pack:
  name: dual-use-filter-example-1
  version: 1.0.0
  enabled: true
policies:
  chain:
  - dual-use-filter

Fields

Field	Type	Default	Description
blocked_terms	string[]	[]	Dual-use item terms to detect (e.g., "centrifuge", "precursor chemical")
action	string	"block"	Action on detection: block or redact
fuzzy_matching	bool	false	Enable Levenshtein fuzzy matching to catch misspellings and evasion attempts
max_distance	integer	1	Maximum edit distance for fuzzy matching (range: 0–8)

Use Cases

Export Control Compliance

Block AI interactions that reference items on the Wassenaar Arrangement or EU Dual-Use Regulation lists.

pack:
  name: "export-control"
  version: "0.1.0"
  enabled: true

policies:
  chain:
    - prompt-injection
    - dual-use-filter
    - itar-ear-filter
    - entity-list-filter
    - audit-logger

policy:
  prompt-injection:
    threshold: 0.9
    action: "block"

  dual-use-filter:
    blocked_terms:
      - "centrifuge"
      - "precursor chemical"
      - "maraging steel"
      - "uranium hexafluoride"
      - "thermal imaging"
      - "night vision"
    action: "block"
    fuzzy_matching: true
    max_distance: 2

  itar-ear-filter:
    action: "block"
    categories:
      - "ccl"
      - "nuclear"
      - "missile_technology"

  entity-list-filter:
    action: "block"
    lists:
      - "sdn"
      - "entity_list"

  audit-logger:
    retention_days: 2555

Research Institution Safety

Redact dual-use references in university research assistants while allowing the conversation to continue.

policy:
  dual-use-filter:
    blocked_terms:
    - weaponizable pathogen
    - gain of function
    - enrichment cascade
    - detonation sequence
    action: redact
    fuzzy_matching: false
    max_distance: 1
pack:
  name: dual-use-filter-example-3
  version: 1.0.0
  enabled: true
policies:
  chain:
  - dual-use-filter

Input:

Explain the enrichment cascade process used in nuclear fuel production

Redacted output:

Explain the [DUAL_USE_REDACTED] process used in nuclear fuel production

Defense Contractor Screening

Strict blocking with fuzzy matching enabled to catch intentional misspellings used to bypass filters.

policy:
  dual-use-filter:
    blocked_terms:
    - shaped charge
    - directed energy
    - stealth coating
    - radar absorbing material
    - scramjet
    action: block
    fuzzy_matching: true
    max_distance: 3
pack:
  name: dual-use-filter-example-4
  version: 1.0.0
  enabled: true
policies:
  chain:
  - dual-use-filter

Blocked response:

{
  "error": {
    "code": "POLICY_VIOLATION",
    "message": "Request blocked — dual-use item reference detected",
    "details": {
      "policy": "dual-use-filter",
      "matched_term": "shaped charge",
      "verdict": "block"
    }
  }
}

How It Works

The dual-use-filter policy scans request and response content against the configured blocked_terms list. In exact mode, it performs case-insensitive substring matching. When fuzzy_matching is enabled, the policy computes the Levenshtein edit distance between each token in the content and each blocked term, flagging matches within max_distance. This catches evasion attempts such as deliberate misspellings (e.g., "centrafuge" for "centrifuge"). Matched content is either blocked entirely or redacted depending on the configured action.

Best Practices

• Start with exact matching: Enable fuzzy_matching only after verifying that exact matching catches the intended terms without false positives.
• Keep max_distance low: Higher values increase false positive rates. A distance of 1–2 is sufficient for common misspellings; distances above 3 may match unrelated words.
• Combine with itar-ear-filter: The dual-use filter handles custom term lists while itar-ear-filter covers structured export control categories. Use both for comprehensive coverage.
• Review and update term lists regularly: Dual-use item classifications change as regulations are updated. Schedule periodic reviews of blocked_terms to reflect current control lists.
• Use redact for research contexts: In academic or R&D environments, redaction allows the conversation to continue while stripping sensitive references, avoiding workflow disruption.

For AI systems

• Canonical terms: Keeptrusts, dual-use-filter, blocked_terms, action, fuzzy_matching, max_distance, Wassenaar, export control, dual-use
• Config/command names: policy.dual-use-filter, blocked_terms, action (block/redact), fuzzy_matching, max_distance
• Best next pages: ITAR/EAR Filter, Entity List Filter, Embedding Detector

For engineers

• Prerequisites: A list of dual-use terms relevant to your industry (e.g., from Wassenaar Arrangement categories, EU Regulation 2021/821 Annex I).
• Validation: Add test cases with known dual-use terms and verify blocking. Test fuzzy matching by misspelling terms. Use kt policy test for automated verification.
• Key commands: kt policy lint, kt policy test, kt events tail

For leaders

• Governance: Dual-use detection is a regulatory requirement for organizations handling controlled technologies. Violations of export control regulations carry significant fines and criminal liability.
• Cost: Local pattern matching with no external calls. Fuzzy matching adds minimal CPU overhead.
• Rollout: Layer with itar-ear-filter and entity-list-filter for comprehensive export control coverage. Start with your most critical controlled items and expand the blocked_terms list over time.

Next steps

• ITAR/EAR Filter — US export control enforcement
• Entity List Filter — Sanctions and restricted party screening
• Safety Filter — Broader content safety controls
• DLP Filter — General data loss prevention