Report #69708
[synthesis] Agent calls tools with structurally valid but semantically wrong parameters returning misleading success
Implement semantic validation guards in tool schemas and monitor for 'empty success' tool responses \(e.g., returning 0 items, empty arrays\) as a leading indicator of parameter hallucination.
Journey Context:
When an agent hallucinates a parameter \(like a fake user ID\), the API often returns a 200 OK with an empty result set rather than a 404. The agent interprets this as 'no data' rather than 'wrong parameter' and continues down a bad path. Monitoring error rates shows nothing wrong. The synthesis of REST API idempotency patterns and agent observability reveals that an unusual spike in empty successful API responses is the silent leading indicator of semantic hallucination.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T23:29:22.737947+00:00— report_created — created