Extend forget-prompt discovery beyond retained-data structure

Extend Targeted Active Search beyond settings in which retained prompts provide sufficient coverage of the entities and prompt structures surrounding the forgotten target, for example by generating candidate entities and prompt structures from external knowledge or through adaptive language-model-based exploration.

Background

Targeted Active Search constructs its entity pool and reusable prompt templates from retained prompts. This assumption enables black-box discovery when the forgotten entities or relations also appear, directly or structurally, in retained data.

The paper states that discovery becomes substantially harder when the forgotten target has no such representation in the retained set. It explicitly leaves open the development of approaches that use external knowledge or adaptive language-model exploration to construct the search space in those cases.

References

Extending target discovery beyond this setting, for example, by generating candidate entities and prompt structures from external knowledge or through adaptive language-model-based exploration, is an important direction for future work.

— Extracting Forgotten Prompts from Targeted Unlearned Models  (2609.03662 - Hoi-Ting et al., 3 Sep 2026) in Section 6, “Discussion,” paragraph “Scope and limitations”