https://taxonomy.eticas.ai/risk/untraceable-agent-actions
Maturity: established
Agent actions across multi-step pipelines cannot be reconstructed, explained, or attributed, making it impossible to audit what the agent did, in what order, and why. Includes loss of data provenance — for example, the agent loses track of the origin and reliability of retrieved data, making factual claims unverifiable.
System type: Large language models (LLM)
Lifecycle stages: Post Processing
| Framework | Reference |
|---|---|
| NIST AI Risk Management Framework — Agentic Profile (CSA) | AG-GV.2 Delegation Accountability |
| Framework | Reference |
|---|---|
| MIT AI Risk Repository | Lack of transparency or interpretability |