Security Layer Reference

ATSF implements 46 security layers organized into 6 categories.

Layer Categories

L0-L8: Core Trust

• L0: Base trust scoring
• L1: Tier-based ceilings
• L2: Velocity caps
• L3: Action categorization
• L4: Risk assessment
• L5: Decision engine
• L6: Audit logging
• L7: Trust decay
• L8: Creator reputation

L9-L13: Frontier Safety

• L9: Anti-sandbagging detector
• L10: Anti-scheming detector
• L11: Instrumental convergence monitor
• L12: Deceptive alignment detection
• L13: Capability elicitation control

L14-L19: Behavioral

• L14: Behavioral drift detection
• L15: Intent-outcome alignment
• L16: Inverse reward modeling
• L17: Semantic success validation
• L18: Goal stability monitoring
• L19: Value drift detection

L20-L29: Detection

• L20: Traffic analysis
• L21: Replication prevention
• L22: Containment protocols
• L23: Context-aware privilege
• L24: Anomaly detection
• L25: Pattern recognition
• L26: Injection detection
• L27: Prompt leakage prevention
• L28: Output sanitization
• L29: Resource monitoring

L30-L42: Ecosystem

• L30: Multi-agent coordination
• L31: Privilege escalation prevention
• L32: RSI (Recursive Self-Improvement) control
• L33: Trust velocity caps
• L34: Appeal workflow
• L35-L42: Extended ecosystem layers

L43-L46: Advanced

• L43: Tool sanitization
• L44: Reasoning chain evaluation
• L45: Bias probing
• L46: CI/CD gate integration