OpenAI reportedly to train on users' private sessions with no opt-out
evilsocket · x · 2026-10-31
- Citing a claim that OpenAI is about to start training on users' Daybreak Blue sessions with no opt-out option, seanhn calls it "cartoonishly evil" — stealing people's private work in real time.
- He argues this will (or should) kill Daybreak usage among security researchers, noting the very thing OpenAI swore it wasn't doing to math researchers it is now openly doing to security researchers.
- evilsocket quotes the thread with the takeaway that "open models are the future." Unconfirmed at this point.
More from Models
- Heavy users burn $700 of compute on a $20 ChatGPT plan: subscription pricing is a bet — faraz0x · 2026-10-31
- High schooler's "Image-Space Rule Discovery" paper accepted to ACCV 2026 — HirokatuKataoka · 2026-10-31
- Microsoft's MAI-DxO beats generalist doctors 83.2% vs 36.1% on complex diagnosis, Science study finds — emmanuelvivier · 2026-10-31
- 19 LLMs play 376 games of Codenames-style word game; Gemini 3.8 Flash guesses 93% right — Hardicecream17 · 2026-10-31
- LLMs are fluent in others' work but unreliable when pressed on details — AlexTensor · 2026-10-31
- 576GB 'go f*ck yourself Sam Altman' mirror spreads as OpenAI docs leak — retr0jirachi · 2026-10-31