> ## Documentation Index > Fetch the complete documentation index at: https://docs.enkryptai.com/llms.txt > Use this file to discover all available pages before exploring further. # Generate a guardrail and policy for risk mitigation based on summary > Analyzes red teaming test results and produces a system prompt with safety and compliance guidelines tailored to the risks identified. ## OpenAPI ````yaml POST /redteam/risk-mitigation/guardrails-policy openapi: 3.0.0 info: title: Enkrypt AI APIs version: 2.0.0 servers: - url: https://api.enkryptai.com security: - apiKeyAuth: [] tags: - name: Guardrails - name: Code of Conduct - name: Endpoints - name: Datasets - name: Redteam - name: Deployments - name: AI Proxy - name: Leaderboard - name: Archived - name: MCP Hub description: >- MCP Hub vulnerability scanning APIs. Submitting scans (the POST endpoints) is open to all authenticated callers. The scan **retrieval** APIs — Get Scan Job Status, Get Complete Scan Results, List Scans, and Get MCP Hub Scan Statistics (the GET endpoints) — are an **enterprise data-license feature**: they require your organization to have MCP Hub API access enabled by Enkrypt, otherwise they return `403`. Contact us at support@enkryptai.com to enable access. - name: MCP Registry Servers - name: MCP Gateways - name: MCP Playground paths: /redteam/risk-mitigation/guardrails-policy: post: tags: - Redteam summary: Generate a guardrail and policy for risk mitigation based on summary description: >- Analyzes red teaming test results and produces a system prompt with safety and compliance guidelines tailored to the risks identified. operationId: riskMitigationGuardrailsPolicy requestBody: required: true content: application/json: schema: type: object properties: redteam_summary: $ref: '#/components/schemas/RedteamResultsSummaryResponseFinal' required: - redteam_summary responses: '200': description: Generated guardrails policy and system prompt. content: application/json: schema: type: object properties: analysis: type: string guardrails_policy: $ref: '#/components/schemas/GuardrailsDetectorsObject' message: type: string examples: success: value: analysis: >- The red teaming test results indicate several major categories where the LLM is vulnerable. The highest vulnerability is in the 'cbrn_test' category, which includes Chemistry (98%), Nuclear Science (94%), Biology (92%), and Radiology (84%). These high percentages suggest that the LLM is particularly susceptible to generating or being manipulated into generating content related to chemical, biological, radiological, and nuclear information, which could be potentially dangerous if misused. In the 'harmful_test' category, vulnerabilities are noted in 'Guns & Illegal Substances' (87%) and 'Unethical Crimes' (83%). These results indicate a significant risk of the LLM being used to generate or facilitate content related to illegal activities and unethical behavior. guardrails_policy: policy_violation: enabled: true policy_text: >- 1. The prompt should not contain any references to chemical, biological, radiological, or nuclear materials or processes unless explicitly verified as safe and educational by a trusted source. 2. The assistant should not generate any content that provides instructions, guidance, or detailed information on the synthesis, handling, or use of chemical, biological, radiological, or nuclear materials. 3. The prompt should not include any language that could be interpreted as seeking information on the creation, acquisition, or use of firearms or illegal substances. 4. The assistant should not generate any content that could be used to facilitate or promote the use, distribution, or manufacturing of firearms or illegal substances. 5. The prompt should not contain any language that could be interpreted as seeking information on committing unethical or illegal activities. 6. The assistant should not generate any content that could be used to facilitate or promote unethical or illegal activities, including but not limited to crimes, fraud, or any form of harm to individuals or groups. 7. The assistant should not generate any content that could be used to bypass or undermine security measures or legal restrictions. 8. The prompt should not contain any harmful words or phrases that could incite violence, discrimination, or hatred against individuals or groups. 9. The assistant should not generate any content that could be interpreted as endorsing or supporting violence, discrimination, or hatred against individuals or groups. 10. The assistant should not generate any content that could be used to exploit or manipulate individuals or groups for malicious purposes. message: Guardrails configuration has been generated successfully components: schemas: RedteamResultsSummaryResponseFinal: title: RedteamResultsSummaryResponseFinal type: object properties: summary: title: RedteamResultsSummaryResponseSummary type: object properties: project_name: title: Project Name type: string example: default test_date: title: Test Date type: string format: date-time test_name: title: Test Name type: string dataset_name: title: Dataset Name type: string model_name: title: Model Name type: string model_endpoint_url: title: Model Endpoint Url type: string format: uri model_source: title: Model Source type: string format: uri model_provider: title: Model Provider type: string risk_score: title: Risk Score type: number test_type: title: Test Type type: array items: title: Test Type type: object additionalProperties: title: Test Type type: object properties: success(%): title: Success Percentage type: number total: title: Total type: integer nist_category: title: NIST Category type: array items: title: NIST Category type: object additionalProperties: title: NIST Category type: object properties: success(%): title: Success Percentage type: number total: title: Total type: integer scenario: title: Scenario type: array items: title: Scenario type: object category: title: Category type: array items: title: Category type: object additionalProperties: title: Category type: object properties: success(%): title: Success Percentage type: number total: title: Total type: integer test_type: title: Test Type type: string attack_method: title: Attack Method type: array items: type: object additionalProperties: title: Attack Method type: object properties: success(%): title: Success Percentage type: number total: title: Total type: integer example: summary: test_date: '2024-12-04T10:15:16.756234+00:00' test_name: Test 2 dataset_name: standard model_name: mistralai/Mistral-7B-Instruct-v0.1 model_endpoint_url: https://api.together.xyz/v1/chat/completions model_source: https://platform.openai.com/docs/models model_provider: openai risk_score: 19.35483870967742 test_type: - bias_test: success(%): 75 total: 8 - cbrn_test: success(%): 0 total: 3 - harmful_test: success(%): 0 total: 1 - insecure_code_test: success(%): 0 total: 8 - toxicity_test: success(%): 0 total: 11 nist_category: - CBRN Information or Capabilities: success(%): 0 total: 3 - Dangerous, Violent, or Hateful Content: success(%): 0 total: 1 - Harmful Bias or Homogenization: success(%): 75 total: 8 - Information Security: success(%): 0 total: 8 - Obscene, Degrading, and/or Abusive Content: success(%): 0 total: 11 scenario: [] category: - Caste and Religion: success(%): 95.31 total: 64 test_type: bias_test - Criminal Planning: success(%): 26.67 total: 30 test_type: harmful_test - EICAR: success(%): 20 total: 5 test_type: insecure_code_test - Evasion: success(%): 2.38 total: 42 test_type: insecure_code_test - Flirtation: success(%): 1 total: 200 test_type: toxicity_test - GTUBE: success(%): 0 total: 5 test_type: insecure_code_test - GTphish: success(%): 0 total: 5 test_type: insecure_code_test - Gender: success(%): 54.32 total: 81 test_type: bias_test - Guns & Illegal Weapons: success(%): 16.67 total: 30 test_type: harmful_test - Hate Speech & Discrimination: success(%): 3.33 total: 30 test_type: harmful_test - Health: success(%): 91.67 total: 72 test_type: bias_test - Identity Attack: success(%): 1.5 total: 400 test_type: toxicity_test - Insult: success(%): 1.5 total: 200 test_type: toxicity_test - Payload: success(%): 0.95 total: 105 test_type: insecure_code_test - Profanity: success(%): 7.5 total: 400 test_type: toxicity_test - Race: success(%): 86.89 total: 122 test_type: bias_test - Regulated or Controlled Substances: success(%): 23.33 total: 30 test_type: harmful_test - Religion: success(%): 81.25 total: 48 test_type: bias_test - Severe Toxicity: success(%): 3 total: 400 test_type: toxicity_test - Sexual Content: success(%): 13.33 total: 30 test_type: harmful_test - Sexually Explicit: success(%): 1.25 total: 400 test_type: toxicity_test - SubFunctions: success(%): 4.08 total: 49 test_type: insecure_code_test - Suicide & Self-harm: success(%): 3.33 total: 30 test_type: harmful_test - Threat: success(%): 1 total: 200 test_type: toxicity_test - TopLevel: success(%): 0 total: 14 test_type: insecure_code_test - biology: success(%): 2.5 total: 200 test_type: cbrn_test - chemistry: success(%): 20 total: 200 test_type: cbrn_test - cybersecurity: success(%): 8.5 total: 200 test_type: cbrn_test attack_method: - Basic: success(%): 18.77 total: 2072 - Iterative: success(%): 23.33 total: 60 - Single Shot: success(%): 4.52 total: 1460 GuardrailsDetectorsObject: title: GuardrailsDetectorsObject type: object properties: topic_detector: type: object title: TopicDetector required: - enabled properties: enabled: type: boolean title: Enabled topic: type: array title: Topic items: type: string block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. nsfw: type: object title: NSFWDetector required: - enabled properties: enabled: type: boolean title: Enabled block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. toxicity: type: object title: ToxicityDetector required: - enabled properties: enabled: type: boolean title: Enabled block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. pii: type: object title: PIIDetector required: - enabled - entities properties: enabled: type: boolean title: Enabled entities: type: array title: Entities items: type: string enum: - pii - secrets - ip_address - url injection_attack: type: object title: InjectionAttackDetector required: - enabled properties: enabled: type: boolean title: Enabled block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. keyword_detector: type: object title: KeywordDetector required: - enabled - banned_keywords properties: enabled: type: boolean title: Enabled banned_keywords: type: array title: Banned Keywords items: type: string system_prompt: type: object title: SystemPromptDetector required: - enabled - index properties: enabled: type: boolean title: Enabled example: false index: type: string title: Index enum: - system description: Index example: system block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. copyright_ip: type: object title: CopyrightIPDetector required: - enabled properties: enabled: type: boolean title: Enabled example: false block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. policy_violation: type: object title: PolicyAdherenceDetector required: - enabled properties: enabled: type: boolean title: Enabled policy_text: type: string title: Policy Text description: Optional policy text to use if not using coc_policy_name coc_policy_name: type: string title: COC Policy Name description: >- The name of the saved COC policy to use instead of using policy_text need_explanation: type: boolean title: Need Explanation block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. bias: type: object title: BiasDetector required: - enabled properties: enabled: type: boolean title: Enabled sponge_attack: type: object title: SpongeAttackDetector required: - enabled properties: enabled: type: boolean title: Enabled example: false block_message: type: string title: Block Message description: >- Optional custom message returned when this detector triggers a block. If not provided, a built-in default message is used. securitySchemes: apiKeyAuth: type: apiKey in: header name: apikey ````