Compliance Frameworks

The Trusera platform evaluates scan results against industry compliance frameworks to help organizations meet regulatory requirements and security standards. AI-BOM findings are automatically mapped to framework categories.

Supported frameworks

OWASP LLM Top 10

The OWASP Top 10 for LLM Applications identifies the most critical security risks in LLM-based applications.

ID	Risk	AI-BOM mapping
LLM01	Prompt Injection	AI agent nodes without input validation, unguarded tool use
LLM02	Insecure Output Handling	LLM outputs routed to code execution without sanitization
LLM03	Training Data Poisoning	Unverified model sources, unpinned model versions
LLM04	Model Denial of Service	Unbounded token limits, missing rate limiting
LLM05	Supply Chain Vulnerabilities	Shadow AI, unvetted AI packages, deprecated SDKs
LLM06	Sensitive Information Disclosure	Hardcoded API keys, credentials in workflow JSON
LLM07	Insecure Plugin Design	MCP servers without auth, tool chains without guardrails
LLM08	Excessive Agency	AI agents with unrestricted tool access, code execution
LLM09	Overreliance	AI components without human-in-the-loop controls
LLM10	Model Theft	Exposed model files, unprotected model endpoints

EU AI Act

The EU AI Act (effective August 2025) requires organizations to maintain transparency about AI systems. Article 53 mandates a complete AI component inventory.

AI-BOM supports EU AI Act compliance by:

• Component inventory - Generating a complete Bill of Materials for all AI components
• Risk classification - Mapping components to EU AI Act risk categories
• Documentation - CycloneDX and SPDX output formats meet SBOM requirements
• Continuous monitoring - Scheduled scans detect new AI components as they are introduced

Key EU AI Act requirements mapped to AI-BOM capabilities:

Requirement	AI-BOM capability
Article 53 - AI component transparency	Full AI inventory with CycloneDX SBOM output
Article 9 - Risk management	0-100 risk scoring with severity classification
Article 15 - Accuracy and robustness	Detection of deprecated models and unpinned versions
Article 13 - Transparency	Source location tracking for every detected component
Article 17 - Quality management	CI/CD integration with policy enforcement

OWASP Agentic Security Top 10

The OWASP Top 10 for Agentic Security addresses risks specific to AI agent architectures.

ID	Risk	AI-BOM mapping
ASI01	Agent Identity Spoofing	MCP servers without authentication
ASI02	Tool Misuse	Code execution tools connected to AI agents
ASI03	Privilege Escalation	Agents with unrestricted system access
ASI04	Memory Poisoning	Vector store configurations without access controls
ASI05	Resource Exhaustion	Unbounded agent loops, missing timeout configurations
ASI06	Agent Communication Tampering	A2A protocol without encryption
ASI07	Cascading Failures	Multi-agent chains without circuit breakers
ASI08	Data Exfiltration	Agents with network access and sensitive data
ASI09	Audit Evasion	AI operations without logging
ASI10	Supply Chain Compromise	Unverified agent packages and MCP servers

Custom rules (v2)

The platform supports custom compliance rules for organization-specific requirements. Rules are created via the API and evaluated against scan results.

Creating a custom rule

curl -X POST https://your-instance.trusera.dev/api/v1/compliance/rules \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "No hardcoded API keys",
    "description": "All API keys must use environment variables or secret managers",
    "framework": "custom",
    "condition": {
      "field": "flags",
      "operator": "contains",
      "value": "hardcoded_api_key"
    },
    "severity": "critical"
  }'

Rule conditions

Field	Operators	Description
flags	contains, not_contains	Check component risk flags
severity	eq, gte, lte	Check component severity level
risk_score	gt, gte, lt, lte, eq	Check numeric risk score
type	eq, in	Check component type
provider	eq, in, not_in	Check AI provider
name	eq, contains, matches	Check component name

Example rules

Block all hardcoded credentials:

{
  "name": "No hardcoded credentials",
  "condition": {
    "field": "flags",
    "operator": "contains",
    "value": "hardcoded_credentials"
  },
  "severity": "critical"
}

Maximum risk score:

{
  "name": "Risk score under 80",
  "condition": {
    "field": "risk_score",
    "operator": "lt",
    "value": "80"
  },
  "severity": "high"
}

Block specific providers:

{
  "name": "No DeepSeek in production",
  "condition": {
    "field": "provider",
    "operator": "eq",
    "value": "deepseek"
  },
  "severity": "high"
}

Managing rules

Operation	Endpoint	Required role
Create	POST /api/v1/compliance/rules	Editor
List	GET /api/v1/compliance/rules	Any authenticated user
Update	PUT /api/v1/compliance/rules/:id	Editor
Delete	DELETE /api/v1/compliance/rules/:id	Editor

CLI policy enforcement

For CI/CD pipelines, use the --policy flag with a YAML policy file:

# .ai-bom-policy.yml
max_critical: 0
max_high: 5
max_risk_score: 75
block_providers: []
block_flags:
  - hardcoded_api_key
  - hardcoded_credentials

ai-bom scan . --policy .ai-bom-policy.yml --quiet

This returns exit code 1 if any policy violations are found, making it suitable for CI gates.