Paper and presentation
Presented orally at SECAI 2026, an ESORICS workshop in Rome in September 2026. The accepted manuscript is available from the authors. Publication in the Springer Lecture Notes in Computer Science proceedings is forthcoming.
Research focus
The paper develops an attack taxonomy and a benchmark for evaluating PII detectors in LLM interactions. It investigates attacks on sensitive values, contextual cues and prompt framing. The project page connects the research findings, code and dataset.
Read the accepted manuscript · Workshop programme · Author project page