Viral app Neon vows to return to sell more of your phone calls to AI companies (www.engadget.com)

Neon — a viral app that pays users for recordings of their phone calls and sells that audio and transcripts to AI companies — went offline after researchers revealed a security flaw that let users access other people’s call recordings, transcripts and metadata. Founder Alex Kiam apologized and told users the service will return “soon” with extra security; Neon also promises users their earnings will be available when it relaunches (it pays up to $30/day: $0.30/min for calls with other Neon users, $0.15/min otherwise, plus $30 referral bonuses). Neon says it records only the caller’s side unless both parties use the app and that automatic filters strip names and phone numbers, but users remain unable to cash out while the outage continues. For the AI/ML community this episode underscores two major issues: the value and legal risk of crowdsourced voice data, and the importance of secure, auditable collection pipelines. Voice recordings are highly prized for training speech and multimodal models, but leaks or poor consent management create regulatory and reputational exposure — especially where two‑party consent laws apply. Automatic PII filtering can help but is imperfect, raising questions about dataset provenance, label quality, bias, and downstream model safety. Companies buying such datasets will face increased pressure to demand verifiable consent, robust access controls, and transparent auditing before integrating third‑party call data into training corpora.