Zen Integration: AI That Follows Your Rules

AI capabilities mean nothing if they violate your policies. Today we are integrating Zen, our rule enforcement system, across all Hanzo AI features.

The Governance Gap

AI systems are powerful. They are also unpredictable. Without governance:

• Chatbots make unauthorized promises
• Recommendations violate compliance rules
• Generated content contradicts brand guidelines
• Agents take actions beyond their authority

The gap between AI capability and policy compliance creates risk.

What is Zen

Zen is a rule enforcement layer that sits between AI capabilities and outputs:

User Request
      ↓
┌─────────────┐
│    Jin      │ ← AI processes request
└──────┬──────┘
       ↓
┌─────────────┐
│    Zen      │ ← Rules evaluated
└──────┬──────┘
       ↓
  Compliant Output

Every AI output passes through Zen before reaching users.

Rule Definition

Rules defined in a declarative language:

# zen-rules.yaml

rules:
  - name: no_competitor_mentions
    description: Never mention competitor brands
    applies_to: [chat, content_generation]
    condition: |
      output.contains_any(competitors.list)
    action: rewrite
    rewrite_instruction: |
      Remove competitor mentions, focus on our products

  - name: price_accuracy
    description: Prices must match catalog
    applies_to: [chat, recommendations]
    condition: |
      output.mentions_price AND
      output.price != product.current_price
    action: block
    fallback: |
      Please check our website for current pricing

  - name: medical_claims
    description: No medical claims for supplements
    applies_to: [content_generation, chat]
    condition: |
      product.category == "supplements" AND
      output.contains_medical_claim
    action: rewrite
    rewrite_instruction: |
      Remove medical claims, use approved language only

Rule Types

Content Rules

Control what AI can say:

- name: brand_voice
  condition: output.tone != "friendly_professional"
  action: rewrite

- name: no_profanity
  condition: output.contains_profanity
  action: block

Data Rules

Control what AI can access:

- name: customer_data_scope
  condition: |
    request.accesses_customer_data AND
    request.customer_id != session.customer_id
  action: block

- name: pii_masking
  condition: output.contains_pii
  action: mask

Action Rules

Control what AI can do:

- name: refund_limit
  applies_to: [support_agent]
  condition: |
    action.type == "issue_refund" AND
    action.amount > 100
  action: require_approval

- name: discount_cap
  condition: |
    action.type == "apply_discount" AND
    action.percent > 20
  action: block

Enforcement Modes

Block

Prevent output entirely, use fallback:

User: "What do you think of [competitor]?"

[Jin generates comparison]
[Zen blocks: competitor mention]

Response: "I'd be happy to help you find the right product
from our catalog. What features are most important to you?"

Rewrite

Modify output to comply:

User: "Tell me about this supplement"

[Jin generates: "This supplement cures headaches..."]
[Zen rewrites: medical claim]

Response: "This supplement is formulated to support overall
wellness. Individual results may vary."

Flag

Allow output but alert for review:

[Output allowed]
[Alert sent to compliance team]
[Logged for audit]

Require Approval

Pause for human approval:

Agent: "I'd like to issue a $150 refund for this customer"

[Zen: requires approval for refunds > $100]

System: "This action requires supervisor approval.
Sending request to @supervisor..."

Audit Trail

Every rule evaluation logged:

{
  "request_id": "req_abc123",
  "timestamp": "2023-09-20T14:30:00Z",
  "rules_evaluated": 12,
  "rules_triggered": 1,
  "triggered_rule": "price_accuracy",
  "action_taken": "block",
  "original_output": "[redacted]",
  "final_output": "Please check our website for current pricing",
  "latency_ms": 8
}

Performance

Zen adds minimal latency:

• Average: 8ms
• p99: 25ms

Rules evaluated in parallel. Common rules cached.

Integration

Zen is now default for all Hanzo AI features. Configure rules in dashboard or via API:

from hanzo import Zen

zen = Zen(api_key="xxx")

# Add rule
zen.rules.create({
    "name": "custom_rule",
    "condition": "...",
    "action": "block"
})

# Evaluate manually
result = zen.evaluate(output, context)

Compliance Templates

Pre-built rule sets for common requirements:

• GDPR: Data access and retention rules
• CCPA: California privacy compliance
• FDA: Supplement and drug claim rules
• FTC: Advertising and endorsement rules
• Financial: Investment advice restrictions

What's Next

• Visual rule builder
• Rule testing sandbox
• Compliance reporting dashboard
• Industry-specific rule packs

AI should be powerful and safe. Zen ensures both.

---

Zach Kelling is the founder of Hanzo Industries.