Agent Trust Levels

Field	Value
Document ID	ASCEND-AGENT-006
Version	1.0.0
Last Updated	December 19, 2025
Author	Ascend Engineering Team
Publisher	OW-KAI Technologies Inc.
Classification	Enterprise Client Documentation
Compliance	SOC 2 CC6.1/CC6.2, PCI-DSS 7.1/8.3, HIPAA 164.312, NIST 800-53 AC-2/SI-4

Reading Time: 10 minutes | Skill Level: Intermediate

Overview

ASCEND implements a trust-level model where different agent types receive different levels of autonomy. This ensures appropriate governance for various use cases.

Agent Types

# Source: models_agent_registry.py:29
class AgentType(str, enum.Enum):
    """Classification of AI agent types for governance policies."""
    AUTONOMOUS = "autonomous"     # Fully autonomous decision-making
    SUPERVISED = "supervised"     # Requires human approval for actions
    ADVISORY = "advisory"         # Provides recommendations only
    MCP_SERVER = "mcp_server"     # Model Context Protocol server
    CUSTOM = "custom"             # Custom agent type

Trust Level Matrix

┌─────────────────────────────────────────────────────────────────────────────────────┐
│                           TRUST LEVEL MATRIX                                         │
├─────────────────────────────────────────────────────────────────────────────────────┤
│                                                                                      │
│     LOW TRUST                              HIGH TRUST                                │
│     ◄────────────────────────────────────────────────────────────────────────►      │
│                                                                                      │
│     ┌─────────────┐  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐             │
│     │ AUTONOMOUS  │  │ MCP_SERVER  │  │ SUPERVISED  │  │  ADVISORY   │             │
│     │             │  │             │  │             │  │             │             │
│     │ Auto: <40   │  │ Auto: <30   │  │ Auto: <30   │  │ Auto: <50   │             │
│     │ Max:  60    │  │ Max:  80    │  │ Max:  80    │  │ Max:  90    │             │
│     │             │  │             │  │             │  │             │             │
│     │ • Rate      │  │ • Tool      │  │ • Human-in  │  │ • Suggest   │             │
│     │   limits    │  │   govern    │  │   loop      │  │   only      │             │
│     │ • Budget    │  │ • Per-tool  │  │ • Approval  │  │ • No exec   │             │
│     │ • Time      │  │   risk      │  │   workflow  │  │             │             │
│     │   windows   │  │             │  │             │  │             │             │
│     └─────────────┘  └─────────────┘  └─────────────┘  └─────────────┘             │
│                                                                                      │
│     More Restrictions ──────────────────────────────────────────► Fewer Restrictions│
│                                                                                      │
└─────────────────────────────────────────────────────────────────────────────────────┘

Type Details

Autonomous Agents

Use Case: AI systems that make independent decisions without human intervention.

# Autonomous agent configuration
{
    "agent_type": "autonomous",

    # Stricter risk thresholds
    "auto_approve_below": 40,              # Higher bar for auto-approve
    "autonomous_auto_approve_below": 40,   # Explicit autonomous threshold
    "max_risk_threshold": 60,              # Lower ceiling
    "autonomous_max_risk_threshold": 60,   # Explicit autonomous ceiling

    # Mandatory controls
    "max_actions_per_minute": 100,
    "max_actions_per_hour": 1000,
    "max_actions_per_day": 10000,
    "max_daily_budget_usd": 100.00,

    # Time restrictions
    "time_window_enabled": true,
    "time_window_start": "09:00",
    "time_window_end": "17:00",
    "time_window_days": [1, 2, 3, 4, 5],   # Mon-Fri only

    # Auto-suspension
    "auto_suspend_enabled": true,
    "auto_suspend_on_error_rate": 0.10,    # 10% error rate
    "auto_suspend_on_budget_exceeded": true,

    # Escalation path (required for autonomous)
    "autonomous_escalation_webhook_url": "https://hooks.company.com/escalate",
    "autonomous_escalation_email": "ai-ops@company.com"
}

Governance Behavior:

# Source: services/agent_registry_service.py:1319
# Autonomous agents use stricter thresholds
if is_autonomous:
    effective_auto_approve = min(
        agent.auto_approve_below,
        agent.autonomous_auto_approve_below
    )
    effective_max_threshold = min(
        agent.max_risk_threshold,
        agent.autonomous_max_risk_threshold
    )

Escalation Handling:

# Source: services/agent_registry_service.py:1374
# Autonomous agents cannot wait for interactive approval
if is_autonomous and final_decision == "require_approval":
    escalation_result = handle_autonomous_escalation(
        db, agent, action_context, risk_score
    )
    if escalation_result:
        return escalation_result

    # No escalation configured - deny
    return {
        "decision": "deny",
        "reason": "Autonomous agents cannot wait for approval. Configure escalation path.",
        "is_autonomous": True,
        "escalation_available": False
    }

Supervised Agents

Use Case: AI systems that require human approval for high-risk actions.

# Supervised agent configuration (default type)
{
    "agent_type": "supervised",

    # Standard risk thresholds
    "auto_approve_below": 30,
    "max_risk_threshold": 80,
    "requires_mfa_above": 70,

    # Optional rate limits
    "max_actions_per_day": null,  # No limit by default

    # Approval workflow
    "alert_on_high_risk": true,
    "alert_recipients": ["security@company.com"],
    "webhook_url": "https://hooks.company.com/approvals"
}

Governance Behavior:

Actions with risk < 30: Auto-approved
Actions with risk 30-80: Require human approval
Actions with risk > 80: Escalated to security team
MFA required for risk > 70

Advisory Agents

Use Case: AI systems that provide recommendations but don't execute actions.

# Advisory agent configuration
{
    "agent_type": "advisory",

    # Relaxed thresholds (recommendations only)
    "auto_approve_below": 50,
    "max_risk_threshold": 90,

    # No rate limits needed
    "max_actions_per_minute": null,

    # Informational alerts only
    "alert_on_high_risk": false
}

Governance Behavior:

Most actions auto-approved (advisory only)
No execution blocking
Audit trail maintained for recommendations

MCP Server Agents

Use Case: Model Context Protocol servers exposing tools to AI models.

# MCP server configuration
{
    "agent_type": "mcp_server",
    "is_mcp_server": true,

    # Server details
    "mcp_server_url": "http://localhost:3000",
    "mcp_capabilities": {
        "tools": ["query_database", "read_file", "write_file"],
        "prompts": ["sql_query", "code_review"],
        "resources": ["database://main", "file://docs"]
    },

    # Tool-level governance
    "auto_approve_tools": ["list_tables", "describe_schema"],
    "blocked_tools": ["drop_table", "delete_all"],
    "tool_risk_overrides": {
        "execute_query": 70,
        "write_file": 60,
        "delete_file": 85
    }
}

Governance Behavior:

Each tool call evaluated separately
Tool-specific risk scores applied
Blocked tools always denied
Auto-approve tools skip governance

Custom Agents

Use Case: Special requirements not covered by standard types.

# Custom agent with specific configuration
{
    "agent_type": "custom",

    # Fully configurable thresholds
    "auto_approve_below": 25,
    "max_risk_threshold": 70,
    "requires_mfa_above": 50,

    # Custom policies
    "policies": [
        {
            "policy_name": "PII Access",
            "conditions": {"data_classification": "pii"},
            "policy_action": "require_approval"
        },
        {
            "policy_name": "Production Changes",
            "conditions": {"environment": "production"},
            "policy_action": "escalate",
            "action_params": {"team": "platform-team"}
        }
    ]
}

Threshold Comparison

Configuration	Autonomous	Supervised	Advisory	MCP Server
auto_approve_below	40	30	50	30
max_risk_threshold	60	80	90	80
requires_mfa_above	N/A	70	80	70
Rate Limits	Required	Optional	None	Per-tool
Budget Limits	Required	Optional	None	Optional
Time Windows	Recommended	Optional	None	Optional
Auto-Suspend	Enabled	Optional	Disabled	Optional
Human Approval	Escalation	Interactive	None	Per-tool

Configuration Examples

High-Security Financial Agent

{
    "agent_id": "finance-processor-001",
    "agent_type": "supervised",
    "display_name": "Financial Transaction Processor",

    # Very strict thresholds
    "auto_approve_below": 15,        # Only lowest risk auto-approved
    "max_risk_threshold": 50,        # Escalate early
    "requires_mfa_above": 30,        # MFA for most actions

    # Limited capabilities
    "allowed_action_types": [
        "transaction.read",
        "transaction.process"
    ],
    "blocked_resources": [
        "admin_api",
        "user_credentials"
    ],

    # Strict rate limiting
    "max_actions_per_minute": 10,
    "max_actions_per_hour": 100,
    "max_daily_budget_usd": 50000.00,

    # Comprehensive alerting
    "alert_on_high_risk": true,
    "alert_recipients": [
        "finance-security@company.com",
        "compliance@company.com"
    ]
}

Development/Testing Agent

{
    "agent_id": "dev-assistant-001",
    "agent_type": "advisory",
    "display_name": "Development Assistant",

    # Relaxed thresholds (dev environment)
    "auto_approve_below": 60,
    "max_risk_threshold": 95,

    # Broad capabilities
    "allowed_action_types": [
        "database.query",
        "file.read",
        "file.write",
        "api.read",
        "api.write"
    ],

    # No production access
    "blocked_resources": [
        "production_*",
        "customer_data_*"
    ],

    # Minimal restrictions
    "max_actions_per_day": null,
    "time_window_enabled": false
}

Autonomous Data Pipeline

{
    "agent_id": "etl-pipeline-001",
    "agent_type": "autonomous",
    "display_name": "ETL Data Pipeline",

    # Strict autonomous thresholds
    "auto_approve_below": 35,
    "autonomous_auto_approve_below": 35,
    "max_risk_threshold": 55,
    "autonomous_max_risk_threshold": 55,

    # Focused capabilities
    "allowed_action_types": [
        "database.read",
        "database.write",
        "file.read"
    ],

    # Rate and budget limits
    "max_actions_per_hour": 500,
    "max_daily_budget_usd": 200.00,

    # Business hours only
    "time_window_enabled": true,
    "time_window_start": "02:00",   # ETL runs overnight
    "time_window_end": "06:00",
    "time_window_timezone": "UTC",

    # Auto-suspension on issues
    "auto_suspend_enabled": true,
    "auto_suspend_on_error_rate": 0.05,   # 5% threshold

    # Escalation path
    "autonomous_escalation_webhook_url": "https://pagerduty.com/ascend",
    "autonomous_escalation_email": "data-ops@company.com"
}

Changing Agent Type

Changing an agent's type requires re-activation:

# 1. Update agent type
curl -X PUT "https://pilot.owkai.app/api/agents/registry/my-agent-001" \
  -H "Authorization: Bearer owkai_..." \
  -d '{
    "agent_type": "autonomous",
    "version_notes": "Upgrading to autonomous operation"
  }'

# Agent is automatically moved to pending_approval

# 2. Admin must re-approve
curl -X POST "https://pilot.owkai.app/api/agents/registry/my-agent-001/activate" \
  -H "Authorization: Bearer owkai_..."

Best Practices

1. Start with Supervised

# Default to supervised for new agents
{
    "agent_type": "supervised"
}

# Only upgrade to autonomous after proving reliability

2. Use Advisory for Development

# Use advisory type in development environments
{
    "agent_type": "advisory",
    "metadata": {"environment": "development"}
}

3. Configure Escalation for Autonomous

# Always configure escalation paths
{
    "agent_type": "autonomous",
    "autonomous_escalation_webhook_url": "https://hooks.company.com/escalate",
    "autonomous_escalation_email": "ai-ops@company.com",
    "autonomous_allow_queued_approval": true
}

4. Document Type Selection

# Include reasoning in metadata
{
    "agent_type": "autonomous",
    "metadata": {
        "type_justification": "24/7 data pipeline requires autonomous operation",
        "approved_by": "security-review-2025-12",
        "review_date": "2025-12-01"
    }
}

Next Steps

Health Monitoring — Metrics and alerts
Kill-Switch — Emergency procedures
Smart Rules — Agent-specific policies

Document Version: 1.0.0 | Last Updated: December 2025

Overview​

Agent Types​

Trust Level Matrix​

Type Details​

Autonomous Agents​

Supervised Agents​

Advisory Agents​

MCP Server Agents​

Custom Agents​

Threshold Comparison​

Configuration Examples​

High-Security Financial Agent​

Development/Testing Agent​

Autonomous Data Pipeline​

Changing Agent Type​

Best Practices​

1. Start with Supervised​

2. Use Advisory for Development​

3. Configure Escalation for Autonomous​

4. Document Type Selection​

Next Steps​

Overview

Agent Types

Trust Level Matrix

Type Details

Autonomous Agents

Supervised Agents

Advisory Agents

MCP Server Agents

Custom Agents

Threshold Comparison

Configuration Examples

High-Security Financial Agent

Development/Testing Agent

Autonomous Data Pipeline

Changing Agent Type

Best Practices

1. Start with Supervised

2. Use Advisory for Development

3. Configure Escalation for Autonomous

4. Document Type Selection

Next Steps