Safety & GovernanceFebruary 14, 202636 min read

Anomaly Detection for Agentic System Safety and Deviation Control

Isolation Forest and Autoencoder reconstruction error as the computational safety layer for self-governing enterprises

Agentic systems can produce operational deviations that require early detection and controlled response. This paper combines Isolation Forest anomaly scoring with Autoencoder reconstruction error to build a layered safety monitor. We define an anomaly-throttle-freeze response cascade and show how the MARIA OS stability guard applies the spectral-radius condition `spectral_radius < 1 - governance_density` in runtime governance.

anomaly-detectionisolation-forestautoencoderdeviation-monitoringrunaway-agentfraud-detectionsafety-layerreconstruction-erroragentic-companyMARIA OS
Industry ApplicationsFebruary 12, 202638 min read

Evidence Coherence Spectral Analysis: Detecting Fraud Through Eigendecomposition of Audit Evidence

Using spectral methods on evidence correlation matrices to identify inconsistencies, fabrication patterns, and systemic fraud signals

Traditional audit systems often rely on rule-based checks and statistical sampling, which can under-detect coordinated fabrication patterns. This paper introduces Evidence Coherence Spectral Analysis, a framework that treats evidence sets as vector spaces, builds correlation matrices from evidence attributes, and applies eigendecomposition to identify anomalous spectral gaps associated with inconsistency or fabrication risk. We define a coherence score, relate it to false-discovery behavior, and describe integration with MARIA OS Evidence Bundles. In controlled financial-statement audit experiments, spectral analysis detected 94.7% of fabricated evidence sets while maintaining a false-positive rate below 2.3%, with streaming support for near-real-time analysis.

auditspectral-analysisevidence-coherencefraud-detectioneigendecompositionmathematicsgovernance