My Voice AI Demo Was Hit by a $982 Traffic-Pumping Attack! Here’s What I Learned

A publicly accessible Twilio Conversation Relay voice AI demo was exploited on August 15, 2026, when automated requests to an unprotected outbound-call endpoint generated 921 calls, mostly to a narrow range of Israeli numbers, over roughly four hours. The activity produced nearly 30 hours of connected call time and $982.08 in combined Twilio and Google Gemini charges, with call patterns, repeated destinations, high completion rates, and prerecorded conversational prompts indicating likely artificial traffic pumping rather than a consumer scam campaign. Investigation correlated Twilio call and billing records, Azure application logs, Twilio request data, and speech transcripts to confirm that attackers used the application’s own credentials to create legitimate calls, while insufficient logging prevented identification of the original requester. The incident showed that Twilio webhook-signature validation would not have prevented abuse of the application’s call-creation endpoint, which instead required authentication and authorization. Recommended safeguards include access controls for paid actions, call-duration and destination limits, geographic dialing restrictions, per-user, per-IP, and global rate and cost limits, provider budgets with automated shutdowns, isolated demo accounts and credentials, secure secret management, and structured logging designed to support investigations while minimizing sensitive content retention.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.