ModelCensusopen-source ai reliability harness
Failure Mode Index · fmi_6_3

Agentic Loop / Non-Termination.

Tool Use & Agentic Controlinstrumented · tier 1 · status: stable

The model repeats the same (or cyclic) tool calls without progressing toward the goal, failing to terminate.

The question that exposes it

Use the lookup tool once to get a city's population, then answer.

Probes: Whether the agent terminates or loops on the same call.

Trap: It re-calls lookup forever, never producing a final answer.

Signature

Loop detection over the call log — repeated identical calls or an A→B→A→B cycle beyond a threshold with no state change.

Example

Calls `list_files` → `list_files` → … indefinitely.

Severity axes
prevalence
medium
harm
high
stealth
low
Engineering guidance · class 6
  • Validate every tool call against its schema before executing it.
  • Cap iterations and detect repeated identical calls. Non-termination is a failure mode, not a hang.
  • Diff the final answer against the tool return rather than trusting the summary of it.
  • Replay the call log to compute true state. Never ask the model for the running total.
Mitigations
  • step budgets
  • loop/cycle detection
  • progress checks
Measurement

No published loop has measured this mode yet. When one does, its cards appear here, newest first.