Search papers, labs, and topics across Lattice.
This paper introduces SPIA, a new benchmark for text anonymization that evaluates the risk of re-identification at the subject level, rather than the span level, using 675 documents from legal and online domains. Their experiments reveal that even with high PII span masking (over 90%), subject-level inference protection can be as low as 33%, indicating significant information leakage through contextual inference. They also find that anonymization strategies focused on a target subject can inadvertently increase the exposure of non-target subjects.
Even when you think you've scrubbed 90% of the PII, your anonymized text might still leak two-thirds of a person's identity.
Current text anonymization evaluation relies on span-based metrics that fail to capture what an adversary could actually infer, and assumes a single data subject, ignoring multi-subject scenarios. To address these limitations, we present SPIA (Subject-level PII Inference Assessment), the first benchmark that shifts the unit of evaluation from text spans to individuals, comprising 675 documents across legal and online domains with novel subject-level protection metrics. Extensive experiments show that even when over 90% of PII spans are masked, subject-level inference protection drops as low as 33%, leaving the majority of personal information recoverable through contextual inference. Furthermore, target-subject-focused anonymization leaves non-target subjects substantially more exposed than the target subject. We show that subject-level inference-based evaluation is essential for ensuring safe text anonymization in real-world settings.