Less is More: Geometric Unlearning for LLMs with Minimal Data Disclosure
For LLM practitioners needing post-hoc privacy removal, GU offers a data-efficient unlearning method that avoids the tension between unlearning strength and utility preservation.
Geometric Unlearning (GU) enables selective unlearning in LLMs without access to original training data, achieving strong target suppression with minimal impact on non-target performance on ToFU and UnlearnPII benchmarks.
As large language models (LLMs) are increasingly deployed in real-world systems, they must support post-hoc removal of specific content to meet privacy and governance requirements. This motivates selective unlearning, which suppresses information about a particular entity or topic while preserving the LLM's general utility. However, most existing LLM unlearning methods require access to the original training corpus and rely on output-level refusal tuning or broad gradient updates, creating a tension among unlearning strength, non-target preservation, and data availability. We propose Geometric Unlearning (GU), an approach that operates directly on the model's prompt-time planning states without access to the original training corpus. GU distills a compact, low-rank geometry of desired safe behavior from a small set of safe reference prompts, and uses lightweight anchor-in-context synthetic prompts to trigger localized, projection-based alignment of hidden planning representations to this safe geometry. A teacher-distillation regularizer on synthetic non-target anchors further reduces collateral drift. Across privacy-oriented unlearning benchmarks (ToFU and UnlearnPII), GU achieves strong target suppression with minimal impact on non-target performance, demonstrating that effective unlearning can be achieved with minimal synthetic data.