A survey of machine unlearning for LLMs covering verification of forgetting, resistance to adversarial recovery, and technical and institutional safeguards, concluding that robust, verifiable unlearning remains unsolved.
A position paper cataloguing how machine unlearning could help comply with the EU AI Act, identifying the technical gaps between current unlearning methods and those compliance uses.
Argues that machine unlearning is not a general-purpose fix for problematic generative-AI content, identifying mismatches between unlearning goals (removing training information or suppressing outputs) and feasible implementations, with policy implications.
Examines how LLMs challenge right-to-be-forgotten compliance and surveys technical remedies such as differential privacy, machine unlearning, model editing, and guardrails.