Observability
Anomaly & Abuse Detection
Anomaly and abuse detection: spot deviant behavior of users, agents and models.
Plane
Observability
Flow steps
9
Frameworks
NIST 800-53 · OWASP LLM10 · NIST AI 600-1
Technology
Why use it
Detect what static rules miss: abnormal usage, abuse, agent behavior drift.
Why it matters to security
Catches unknown attacks and slow abuse (quiet exfiltration, abnormal consumption) no signature covers.
Implementations UEBAML anomaly detectioncost/consumption analyticsFalco
Novel attacks match no signature: you must detect the anomaly, not only the known.
Recommendations by maturity tier
Foundation
Minimum viable baseline
- Monitoring of key indicators (throughput, errors, cost). NIST 800-53 SI-4Basic metrics already reveal a lot of abuse.
- Alert thresholds on abnormal consumption. NIST 800-53 SI-4OWASP LLM10:2025Out-of-norm consumption signals abuse or runaway loops.
- Regular log review. NIST 800-53 AU-6Human analysis complements automated detection.
Enterprise
Enterprise standard
- Behavioral baselines. NIST 800-53 SI-4Compare current behavior to a learned norm.
- AI-specific abuse detection (repeated jailbreak). NIST AI 600-1 MS-2.7-009Repeated bypass attempts are spotted.
- Risk scoring feeding the PIP. NIST 800-53 RA-3The anomaly score becomes a decision signal for the PDP.
Advanced
High-assurance / regulated
- Adaptive behavioral detection (UEBA). NIST 800-53 SI-4The detection model continuously learns new usage.
- Multi-signal identity/data/agent correlation. NIST 800-53 SI-4 · AU-6A cluster of weak signals reveals a sophisticated attack.
- Remediation loop wired to SOAR. NIST AI 600-1 MG-4.1-002A confirmed anomaly triggers an automated response.
Architecture notes
- Watch cost as a security signal.details ▸A consumption drift often betrays abuse or an agent loop.Feed cost/token metrics into anomaly detection, not just financial reporting.
References
NIST SP 800-53 Rev5
SI-4 (System Monitoring), AU-6 (Analysis), RA-3 (Risk Assessment).
OWASP LLM10:2025
Unbounded Consumption — abuse detection is a key facet.
NIST AI 600-1
MS-2.7-009 (attack tracking), MG-4.1 (remediation).
Abbreviations
PDP
Policy Decision Point
PEP
Policy Enforcement Point
PIP
Policy Information Point
PAP
Policy Administration Point
IdP
Identity Provider
TSS
Token Service
NHI
Non-Human Identity
RBAC
Role-Based Access Control
ABAC
Attribute-Based Access Control
MFA
Multi-Factor Authentication
HITL
Human-in-the-loop
JIT
Just-In-Time
CAE
Continuous Access Evaluation
CAEP
Continuous Access Evaluation Profile
DPoP
Demonstrating Proof-of-Possession
mTLS
mutual TLS
PII
Personally Identifiable Information
KMS
Key Management Service
CI/CD
Continuous Integration / Continuous Delivery
SIEM
Security Information and Event Management
SOAR
Security Orchestration, Automation and Response
SCIM
System for Cross-domain Identity Management
XACML
eXtensible Access Control Markup Language
OPA
Open Policy Agent
OWASP
Open Worldwide Application Security Project
NIST
National Institute of Standards and Technology
ATLAS
Adversarial Threat Landscape for Artificial-Intelligence Systems
LLM
Large Language Model
WAF
Web Application Firewall
CDN
Content Delivery Network
DDoS
Distributed Denial of Service
DLP
Data Loss Prevention
JWT
JSON Web Token
API
Application Programming Interface
CRS
Core Rule Set (OWASP)
RAG
Retrieval-Augmented Generation
MCP
Model Context Protocol
PBAC
Permission-Based Access Control
HSM
Hardware Security Module
UEBA
User and Entity Behavior Analytics
SBOM
Software Bill of Materials
SLSA
Supply-chain Levels for Software Artifacts
WORM
Write Once, Read Many
SPIFFE
Secure Production Identity Framework For Everyone