Feature
Policy Enforcement
Define governance rules in human-readable YAML and enforce them consistently across all your AI operations. Block, warn, redact, or allow based on content, user context, or custom conditions.
How It Works
Tork's Policy Engine evaluates every AI input and output against your defined rules before they're processed or returned. This happens in milliseconds, adding negligible latency while providing complete governance control.
policy.yaml
policies:
# Block harmful content
- name: content-safety
trigger: output
action: BLOCK
conditions:
- type: contains_harmful_content
threshold: 0.8
message: "Content blocked due to safety policy"
# Redact PII before sending to AI
- name: pii-protection
trigger: input
action: REDACT
patterns:
- email
- phone
- ssn
- credit_card
# Warn on sensitive topics
- name: sensitive-topics
trigger: output
action: WARN
keywords:
- medical_advice
- legal_advice
- financial_advice
escalate_to: compliance@company.comPolicy Actions
Let the content pass through unchanged. Use for approved patterns or trusted sources.
Reject the request entirely. Returns a configurable error message to the user.
Remove or mask sensitive content while allowing the request to proceed.
Allow but flag for review. Optionally notify compliance teams via webhook.
Integration Example
Python
from tork import PolicyEngine
# Load your policies
engine = PolicyEngine.from_yaml("policy.yaml")
# Evaluate input before sending to AI
user_input = "Contact me at john@example.com"
input_result = engine.evaluate(user_input, trigger="input")
if input_result.action == "BLOCK":
return {"error": input_result.message}
# Safe input - proceed with AI call
ai_response = call_your_ai_model(input_result.content)
# Evaluate output before returning to user
output_result = engine.evaluate(ai_response, trigger="output")
if output_result.action == "BLOCK":
return {"error": "Response blocked by policy"}
# Return governed response
return {"response": output_result.content}Use Cases
Healthcare AI
Block medical diagnoses, ensure HIPAA compliance, redact patient identifiers
Financial Services
Prevent unauthorized financial advice, comply with SEC regulations
Customer Support
Ensure brand-safe responses, block competitor mentions, enforce tone guidelines
Education
Age-appropriate content filtering, prevent academic dishonesty patterns
Legal Tech
Block unauthorized legal advice, ensure jurisdiction-appropriate responses
Why Policy-as-Code?
Version Controlled
Track policy changes in Git. Review, approve, and rollback like code.
Environment Parity
Same policies in dev, staging, and production. No configuration drift.
Auditable
Every policy change is logged. Know who changed what and when.
Testable
Unit test your policies before deployment. CI/CD integration ready.
Declarative
Define what you want, not how. Tork handles the implementation.
Portable
Move between frameworks without rewriting rules.