TRACE reconstructs unlearned LLM text from pre- and post-unlearning snapshots
The attack needs no data-specific prior and was tested on TOFU, MUSE, and WMDP with six unlearning methods.
Stories per day · 30 days
2 stories, newest first · one card per story
Stories tagged #text-reconstruction
LLM-Attacker standardizes privacy tests for split language models
In a GPT-2 test, returned gradients raised token recovery from 94.20% to 97.38% and exact-document recovery from 13.71% to 37.77%.