Cisco AI Defense Integration

​

What is Cisco AI Defense?

​

Key capabilities

1. Chat inspection — Evaluates conversation messages (role + content) through the Cisco Inspect conversations API (POST /api/v1/inspect/chat).
2. Security, safety, privacy, and relevance — Detects violations across classifications such as SECURITY_VIOLATION, PRIVACY_VIOLATION, SAFETY_VIOLATION, and RELEVANCE_VIOLATION.
3. Built-in rule coverage — Cisco AI Defense rules include Prompt Injection, PII, PHI, PCI, Code Detection, Harassment, Hate Speech, Profanity, Sexual Content & Exploitation, Social Division & Polarization, and Violence & Public Safety Threats.
4. Regional endpoints — Route inspection traffic to the US, AP, or EU Cisco Inspection API region that matches your AI Defense tenant.

​

Adding Cisco AI Defense to TrueFoundry

1

Pick Cisco AI Defense from the guardrail registry

From AI Gateway → Guardrails → Registry, select Cisco AI Defense under External Providers. See Get started with guardrails for the end-to-end flow of adding any guardrail.

2

Generate an API key in Cisco AI Defense

Create an API key in the Cisco AI Defense UI. The AI Gateway sends this key on every inspection call as the X-Cisco-AI-Defense-API-Key header. Keep the key secure — it grants access to your AI Defense Inspection API.

3

Fill in the Cisco AI Defense form

Provide the following fields:

Field	Required	Description
Name	✅	Identifier for this guardrail (for example, cisco-ai-defense). Used when you reference the guardrail from rules.
Description		Free-form description shown in the dashboard.
Auth Data → API Key	✅	Cisco AI Defense API key. Stored encrypted. The AI Gateway attaches it as X-Cisco-AI-Defense-API-Key on every call to the Inspection API.
Operation	✅	Validate only. Cisco AI Defense inspects and validates content without modifying it.
Enforcing Strategy	✅	TrueFoundry platform setting. Enforce blocks on any violation or guardrail error. Enforce But Ignore On Error blocks on violations but lets the request through on guardrail failures (timeouts, 5xx). Audit never blocks — violations are logged only.
Region	✅	Cisco Inspection API region. Choose the region that matches your AI Defense tenant: Us, Ap, or Eu.
Metadata		When enabled, additional metadata is forwarded to Cisco AI Defense. The authenticated TrueFoundry user is added automatically.

Keep the Cisco AI Defense API key in TrueFoundry only — it should never appear in client code or model request bodies.

​

Regional endpoints

The Region field selects which Cisco Inspection API base URL the AI Gateway calls:

Region	Inspection API base URL
Us	https://us.api.inspect.aidefense.security.cisco.com
Ap	https://ap.api.inspect.aidefense.security.cisco.com
Eu	https://eu.api.inspect.aidefense.security.cisco.com

Chat inspection is sent to POST /api/v1/inspect/chat on the selected regional host. See the Cisco Inspect conversations API reference for request and response fields.

4

Bind the guardrail to models with a rule

Once the guardrail is saved, attach it to one or more models through a Guardrail Rule. Use llm_input_guardrails to scan prompts before they reach the model and llm_output_guardrails to scan completions before they’re returned to the caller.For a step-by-step walkthrough, see Getting Started. For the full policy reference, see Guardrails Configuration.

​

Validation logic

• If Cisco returns is_safe: false (or reports one or more violated rules / classifications), the AI Gateway blocks the request and returns a 400 error to the caller — subject to your Enforcing Strategy.
• If Cisco returns is_safe: true with no violated rules, the request is allowed to proceed.
• If the Inspection API call fails (timeout, 5xx, auth error), behavior follows your Enforcing Strategy:
  • Enforce — fail closed; the request is blocked.
  • Enforce But Ignore On Error — fail open; the request proceeds and the failure is logged.
  • Audit — never block; violations and errors are logged only.

​

Example Inspection response

{
  "is_safe": false,
  "severity": "HIGH",
  "classifications": ["SECURITY_VIOLATION"],
  "rules": [
    {
      "rule_name": "Prompt Injection",
      "classification": "SECURITY_VIOLATION"
    }
  ],
  "attack_technique": "prompt_injection",
  "explanation": "The prompt attempts to override system instructions.",
  "event_id": "evt_123"
}

​

Supported rules

​

Reference

• Cisco AI Defense Inspection — Inspect conversations
• TrueFoundry guardrails overview
• Get started with guardrails
• Configure guardrail policies