Consent to training data isn't consent to AI racing you to your own result, say mathematicians

PMinervini · x · 2026-09-09

Mathematician Yoav Goldberg argues that "I'm ok with models being trained on my data and released publicly" is fundamentally different from "I'm ok with internal models trained on my very recent efforts racing the solution before me, after rumors I'm close to it." Boaz Barak adds that users can opt out of (de-identified) data use in training and notes Anthropic has similar policies, questioning why such discussions didn't arise when Anthropic announced mathematical results.

This is part of the same math-community dispute over AI training-data ethics and attribution.

Related event: OpenAI's default training on user chats sparks race-to-publish controversy(6 posts)→

Original post →

More from AGI Musings

AGI Musings channel →