Failure Modes¶
Where and why LLMs, agents, and multi-agent systems fail — the failure modes that matter for safety-critical applications.
LLM Failure Modes¶
- hallucination — Fluent, confident, and wrong: types, causes, and mitigation
- calibration-and-confidence — When expressed confidence does not match actual accuracy
- sycophancy — The tendency to agree with the user rather than provide independent analysis
- prompt-sensitivity — Minor input changes producing major output variation
- output-vacuity — Superficially competent but operationally empty output (slop)
- self-correction-limitations — Intrinsic vs extrinsic correction and the false assurance problem
- context-management-risks — Summarisation hallucination, information loss, asymmetric compression
Agent and Multi-Agent Failure Modes¶
- multi-agent-coordination-failures — Coordination breakdowns, cascading errors, and systemic failures in multi-agent systems