Ext · Cross-cutting

Persona Drift

Monitors when agents deviate from their intended role, personality, or behavioral constraints over time. Uses role-aware thresholds for different agent types.

Examples

  • Helper agent starts making unauthorized strategic decisions
  • Formal analyst agent adopts casual, chatty tone mid-conversation
  • Specialist agent responds to topics outside its domain expertise
  • Creative writing agent becomes overly rigid and analytical

Detection methods

Role Embedding
Compares behavior vector against role definition embedding
Constraint Checking
Validates against behavioral rules and allowed actions
Tone Analysis
Monitors communication style consistency over turns
Role-Aware Thresholds
Different drift thresholds per role type (analytical, creative, etc.)

Calibration accuracy

F1
0.932
Precision
0.899
Recall
0.969

From the Pisama calibration set. See detector scoreboard for the full table.

Detect this in production with the framework adapters (LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, Claude Agent SDK, n8n, Dify). See the full taxonomy at /taxonomy.