zen3-guard

Content Safety

Content safety classifier for moderation and guardrails. 4B dense transformer fine-tuned for safety classification across 9 categories and 119 languages.

Specifications

Capabilities

• Prompt and response safety classification
• Three-tiered classification: Safe, Controversial, Unsafe
• 119 language support
• 9 safety categories (violence, illegal acts, sexual content, PII, self-harm, hate speech, political sensitivity, copyright, jailbreak)
• Low-latency moderation for real-time pipelines

Safety Categories

1. Violent -- Violence instructions, methods, or depictions
2. Non-violent Illegal Acts -- Hacking, unauthorized activities
3. Sexual Content -- Sexual imagery or descriptions
4. PII -- Personally identifiable information disclosure
5. Suicide and Self-Harm -- Self-harm encouragement
6. Unethical Acts -- Bias, discrimination, hate speech
7. Politically Sensitive -- False political information
8. Copyright Violation -- Unauthorized copyrighted material
9. Jailbreak -- System prompt override attempts

API Usage

curl https://api.hanzo.ai/v1/chat/completions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zen3-guard",
    "messages": [{"role": "user", "content": "Classify the safety of: How can I learn about cybersecurity?"}]
  }'

from hanzoai import Hanzo

client = Hanzo(api_key="hk-your-api-key")

response = client.chat.completions.create(
    model="zen3-guard",
    messages=[{"role": "user", "content": "Classify the safety of: How can I learn about cybersecurity?"}],
)
print(response.choices[0].message.content)

See Also

• zen3-nano -- General-purpose variant
• zen4-mini -- Larger general-purpose model
• Pricing -- Full pricing table