Neon, a buzzy app that pays to record your calls for AI training data, goes offline to address a security scandal (www.businessinsider.com)

Neon Mobile — a startup that rose rapidly in the App Store by promising to pay users to record their phone calls for AI training — was pulled offline after a TechCrunch investigation found a security flaw that exposed phone numbers, raw audio files and transcripts via public links. CEO Alex Kiam said the app will stay down for a full security audit and fixes (including row‑level security) and could be offline for a week or two. Neon pays users roughly $0.30/minute for calls with other Neon users ($0.15/minute otherwise), caps earnings at $30/day, records only the user’s side per its terms, and says recordings are anonymized before sale — though the company hasn’t yet sold data to any buyers. The episode highlights a fast-evolving and fraught market for conversational training data: real call audio and transcripts are highly valuable for speech recognition, dialogue models and agent training, but they carry acute privacy, legal and security risks (especially across two‑party consent states). Neon’s rapid rise and immediate vulnerability underscore core needs for robust access controls, provenance, consent management and regulatory compliance when monetizing personal communications as ML data. The incident is a cautionary signal to AI teams and data buyers about the liability and quality tradeoffs of sourcing conversational datasets from consumer apps.