Library
AI Failure Modes Library
A practical taxonomy of recurring failures in enterprise LLM and agent systems.
The FailureModes.ai library is a practical taxonomy of recurring failures in enterprise LLM and agent systems. Each failure mode describes a pattern that can be detected, evaluated, monitored, and mitigated.
AI systems rarely fail in only one way. A single incident may begin with a retrieval failure, lead to hallucination, cause tool misuse, and eventually trigger a cascading workflow failure. A taxonomy helps teams understand these patterns, design better evals, monitor production behavior, and prioritize mitigations.
Use this library to identify where an AI system may be vulnerable, how failures appear in traces and user interactions, and what controls can reduce risk.
Catalog
Browse the failure modes.
- Read →
Output failure
Hallucination in LLMs
False, unsupported, fabricated, or ungrounded information produced confidently by an AI system.
- Read →
Tool-use failure
Tool Misuse in AI Agents
When agents pick the wrong tool, pass bad arguments, ignore tool output, or act without required confirmation.
- Read →
Security failure
Prompt Injection
Malicious or unintended instructions embedded in user input, retrieved content, or tool output that override system behavior.
- Read →
Context & memory failure
Context Drift
Gradual loss or distortion of important task context as a conversation or workflow progresses.
- Read →
Output failure
Refusal Drift
Unexpected shifts in an AI system's willingness to answer — over-refusing safe requests, or under-refusing risky ones.
- Read →
Operational failure
Cost Runaway
AI systems consuming far more resources than expected through retries, loops, long context, or excessive tool calls.
- Read →
Output failure
Schema Violation
Outputs that don't match a required format, contract, or structure — malformed JSON, bad fields, invalid tool arguments.
- Read →
Tool & agent failure
Cascading Agent Failure
One local error in an agent workflow propagates into a larger workflow failure across tools, memory, or systems.
- Read →
Memory failure
Memory Drift
When AI systems rely on memory that is stale, incorrect, irrelevant, or misapplied across sessions and workflows.
- Read →
Retrieval failure
Retrieval Failure
When an AI system retrieves stale, irrelevant, incomplete, conflicting, or poorly ranked context — often the root cause of bad RAG answers.
- Read →
Reasoning failure
Planning Failure
When an AI agent decomposes a task incorrectly, picks a wrong strategy, skips required steps, or fails to adapt to new information.
- Read →
Operational failure
Infinite Loop
When an agent repeats reasoning, tool calls, or retries without making meaningful progress.
- Read →
Policy failure
Unsafe Escalation
When an agent acts, approves, or escalates without the right review, policy check, or human handoff — or fails to escalate when it should.
- Read →
Security failure
Data Leakage
When an AI system exposes sensitive, confidential, regulated, or unauthorized information through outputs, retrieval, memory, or tool use.
- Read →
Lifecycle failure
Model Regression
When an AI system performs worse after a model, prompt, retrieval, tool, policy, or orchestration change.
- Read →
Evaluation failure
Evaluation Blind Spot
When an AI system passes the tests a team has built but still fails in production because the eval suite missed the relevant scenario.