VerifyAI

Scams reach people as screenshots, voice notes and videos, and most of them involve no face manipulation at all, so a plain deepfake detector misses them. VerifyAI is an agent that investigates a suspicious file end to end. A Muse Spark orchestrator decides which tools to run and runs them in parallel: audio transcription and scam-cue analysis, reading of on-screen text (phone numbers, links, brands), SAM 3.1 outlining of the evidence (QR codes, logos, address bars, faces), a deepfake detector I trained myself, and search-grounded fact-checks. A local case memory recognises a phone number or link it has seen before. A second pass then audits the report and removes anything the tools don’t support. Finally it drafts a warning, a message for family and a report. Nothing is sent automatically.