Julep, an AI agent platform, contains an insecure direct object reference (IDOR) vulnerability in its get_execution_details endpoint. Authenticated tenants can exploit this flaw by supplying arbitrary execution_id values to access execution data belonging to other tenants. The exposed data includes sensitive task inputs, outputs, metadata, and temporal task tokens. This represents a significant multi-tenant isolation failure allowing cross-tenant data leakage. The vulnerability is tracked as CVE-2026-67348 and has been reported via GitHub issues and documented by VulnCheck. The issue affects the authorization layer of the API, which fails to validate that the requested execution_id belongs to the requesting tenant. Exploitation requires only valid authentication credentials, lowering the barrier for abuse.