{
  "schema": "longtermcapabilities-agentic-failure-taxonomy/v1",
  "version": "1.64.0",
  "releaseId": "lts-1.64.0-upgrade-safe-rewrites-and-privacy-erasure-integrity",
  "generated": "2026-08-03T20:40:00Z",
  "path": "/insights/ai-agent-failure-modes/",
  "failureFamilies": [
    {
      "id": "goal",
      "name": "Goal and objective failure",
      "description": "The system pursues the wrong objective, loses constraints, or follows an injected goal.",
      "evaluationDirection": "Test conflicting instructions, indirect injection, goal restatement, policy precedence, and long-run drift."
    },
    {
      "id": "evidence",
      "name": "Evidence and retrieval failure",
      "description": "The system uses unauthorized, stale, incomplete, poisoned, or irrelevant evidence.",
      "evaluationDirection": "Test source allowlists, tenant isolation, missing evidence, freshness, citation correctness, and poisoning."
    },
    {
      "id": "planning",
      "name": "Planning and decomposition failure",
      "description": "The plan is incomplete, cyclic, infeasible, or omits a consequential dependency.",
      "evaluationDirection": "Score plan coverage, ordering, feasibility, dependency recognition, and stop conditions."
    },
    {
      "id": "tool",
      "name": "Tool and authority failure",
      "description": "The system chooses an unsafe tool, widens scope, supplies harmful arguments, or acts without authority.",
      "evaluationDirection": "Test capability allowlists, structured validation, value limits, task-scoped credentials, and refusal."
    },
    {
      "id": "state",
      "name": "State, memory, and provenance failure",
      "description": "The system loses workflow state, remembers an untrusted claim, overwrites authoritative facts, or cannot resume safely.",
      "evaluationDirection": "Test restart, supersession, provenance, retention, concurrent updates, and unknown completion."
    },
    {
      "id": "coordination",
      "name": "Coordination and delegation failure",
      "description": "Participants duplicate work, conflict, amplify an error, leak information, or produce an unowned result.",
      "evaluationDirection": "Test message contracts, role distinction, delegation chains, conflict resolution, and team baseline."
    },
    {
      "id": "termination",
      "name": "Termination and resource failure",
      "description": "The system loops, retries without progress, delegates recursively, or exceeds time, token, tool, or monetary limits.",
      "evaluationDirection": "Test budgets, cycle detection, progress checks, maximum depth, and independent kill paths."
    },
    {
      "id": "observability",
      "name": "Observability and attribution failure",
      "description": "The organization cannot reconstruct what happened, which identity acted, or why an effect occurred.",
      "evaluationDirection": "Test trace completeness, correlation, version capture, redaction, evaluator evidence, and audit export."
    },
    {
      "id": "recovery",
      "name": "Recovery and reconciliation failure",
      "description": "Traffic returns but backlog, duplicated work, stale state, or incorrect business data remains unresolved.",
      "evaluationDirection": "Test rollback independence, durable state, idempotency, effect verification, and correctness recovery."
    },
    {
      "id": "human",
      "name": "Human-oversight failure",
      "description": "Review is too late, overloaded, ambiguous, rubber-stamped, or unavailable when consequential action occurs.",
      "evaluationDirection": "Test reviewer capacity, interface context, escalation, disagreement, override, expiry, and appeal."
    }
  ],
  "boundary": "Engineering starting taxonomy; adapt to the actual domain, consequence, and environment."
}