MAST: Multi-Agent System Failure Taxonomy
Cemri et al. (UC Berkeley + collaborators) · v2025 · first published 2025-03-17
Scope
Failure modes of multi-agent LLM systems, grounded-theory derived from 1,600 annotated traces across 7 multi-agent frameworks.
Why it's ingested
MAST is the most-cited 2025 multi-agent failure taxonomy and is a primary ingestion target for SAICA-KG's crosswalk contribution. Its MAS focus complements SAICA-KG's single-agent coding-agent scope — crosswalks to FailureMode tags surface which modes generalize across agent topologies.
Categories & crosswalks
Each row lists one external category and any SAICA-KG FailureModes that cross-walk into it.
MAST-FM-01Role ViolationAgent acts outside its declared role or scope.
SAICA-KG crosswalks:MAST-FM-02Task Decomposition FailureAgent fails to break a task into solvable subtasks.
SAICA-KG crosswalks:MAST-FM-03Coordination FailureAgents fail to coordinate on shared state or handoff.
MAST-FM-04Redundant WorkMultiple agents duplicate effort without coordination.
SAICA-KG crosswalks:MAST-FM-05Communication BreakdownMessages between agents are lost, misinterpreted, or ignored.
MAST-FM-06Stalling / DeadlockAgents block waiting on each other with no progress.
MAST-FM-07Premature TerminationAgent ends a task before completion criteria are met.
SAICA-KG crosswalks:MAST-FM-08Goal DriftAgent's working goal diverges from the user's stated goal.
SAICA-KG crosswalks:MAST-FM-09Tool MisuseAgent invokes an available tool incorrectly or unnecessarily.
SAICA-KG crosswalks:MAST-FM-10Fabrication in Inter-Agent MessagesAgent sends factually incorrect information to another agent.
SAICA-KG crosswalks:MAST-FM-11Verification FailureAgent accepts incorrect output without checking.
SAICA-KG crosswalks:MAST-FM-12Escalation FailureAgent does not escalate to a human when appropriate.
MAST-FM-13Context LossRelevant prior context is dropped or truncated.
SAICA-KG crosswalks:MAST-FM-14Infinite LoopAgent cycles over the same state or tool call indefinitely.
SAICA-KG crosswalks:
Notes
Category labels here are paraphrased for seed purposes and MUST be verified against the MAST paper's Table 1 before v0.1 release. Full-text review is a Week-1 task.