Nearly all AI models' ToS allow training on your data, researcher notes amid Codex privacy debate

_arohan_ · x · 2026-09-08

Amid debate over whether user data influenced a model behavior, arohan says it's unlikely OpenAI pulled user transcripts, but points out nearly all AI models' ToS allow training on user data, with opt-out details unclear. Building big-tech-grade privacy isolation takes massive infra, which is why many push for ZDR.

Related event: OpenAI Accused of Scooping Navier-Stokes Work; Bubeck Denies Allegations(35 posts)→

Original post →

More from Models

Models channel →