Crypto researcher Green: multiple scientists suspect AI models train on their inputs

matthew_d_green · x · 2026-09-09

Matthew Green, cryptographer at Johns Hopkins, said multiple researchers in his field have noticed models getting much better at solving their specific problems over short periods—even when asked in new contexts. Several have wondered whether the models are being trained on their inputs. The claim remains anecdotal and unconfirmed, but touches the sensitive question of whether user data flows into model training.

Original post →

Crypto researcher Green: multiple scientists suspect AI models train on their inputs

More from Models