AI agent hallucinations raise reliability risks in booking and identity workflows

Reliability concerns are resurfacing for AI agents: hallucinations in knowledge retrieval were perceived to have receded but are now described as returning. Frontier tasks such as system design, Blender and AI video remain more error-prone.

In one cited flight-booking case, an agent reportedly generated the middle name “Bom Chang Choi” for David, potentially disrupting boarding. The error may be corrected through airline support and a new boarding pass, but it could still delay security or cause a missed flight.

One explanation offered is cost pressure: free products served at scale may use older models or other cost-saving approaches. A cloud-hosted OpenClaw instance was tentatively identified as possibly running Kimi; that attribution remains uncertain.