Machine-intelligence working framework

Multi-Agent Cascading Failure Taxonomy

Classify where a cascade begins, how it crosses a boundary, what amplifies it, and which control can interrupt it before side effects spread.

EnterpriseGovernmentPartners

Published August 1, 2026Updated August 1, 2026Reviewed August 1, 2026By LongTermIntelligence.com

Direct answer

What is a cascading failure in a multi-agent system?

A cascading failure occurs when a local error, unsupported assumption, stale state, or unauthorized action is accepted by another agent or tool and amplified across the workflow. The important unit of analysis is the boundary crossing, not only the original model output.

A technically valid handoff can still be semantically wrong.
Shared memory can preserve and amplify a bad state.
Containment requires boundaries, validation, budgets, and rollback.

Source basis: reviewed synthesis of the strategy corpus. Report-derived claims remain subject to the verification boundary in the source library.

Failure taxonomy

Six cascade classes

Classify the failure before selecting a mitigation.

F-01

Boundary mismatch

The receiver interprets a field, intent, confidence, or completion state differently from the sender.

F-02

Stale or conflicting state

Agents act on different versions of the task, world state, policy, or prior decision.

F-03

Authority bypass

Delegation or tool composition produces an action no single component was authorized to perform.

F-04

Context contamination

Unsupported, malicious, irrelevant, or cross-session material enters memory or retrieval and persists.

F-05

Coordination deadlock or storm

Agents wait indefinitely, duplicate work, loop, or trigger uncontrolled retries and calls.

F-06

Side-effect amplification

A bad intermediate result mutates systems, commits resources, or changes downstream decisions before review.

Containment sequence

Interrupt the cascade at the earliest defensible point

The response should preserve enough evidence to reconstruct what happened.

Detect
Watch boundary contracts, state versions, policy decisions, budgets, and outcome invariants.
Quarantine
Suspend the affected task, agent identity, tool, memory namespace, or downstream action.
Preserve
Retain versions, traces, messages, tool calls, approvals, and external state.
Recover
Compensate or roll back side effects, then resume from a trusted checkpoint.
Learn
Add the failure to evaluation suites, update controls, and review authority and ownership.

Downloadable working files

Use the risk register to operationalize the taxonomy

CSV template

Production Agent Risk Register

Record each cascade class as a scenario with controls, signals, owners, and response.

Download CSV

Templates are planning aids. They are not certifications, legal advice, security guarantees, or substitutes for client-specific validation.

Next decision

Design for the cascade, not only the component

Run a tabletop exercise around one high-consequence handoff before production release.

Discuss the decision Evaluate readiness

Multi-Agent Cascading Failure Taxonomy

What is a cascading failure in a multi-agent system?

Boundary mismatch

Stale or conflicting state

Authority bypass

Context contamination

Coordination deadlock or storm

Side-effect amplification

Interrupt the cascade at the earliest defensible point

Detect

Quarantine

Preserve

Recover

Learn

Production Agent Risk Register

Continue through the knowledge system

Agentic incident response

Evaluation Scorecard

Agents fail at boundaries

Design for the cascade, not only the component