Defense suite for persona skills: online sanitization, post-hoc adversarial obfuscation, and semantic backdoor watermarking¶
Strategy · All strategies
Evidence · 2 claims (2 for) · 1 study (1 design),q2· 0 of 1 report an effect size · 2 claims rest on one study
Description¶
AntiSkillBench designs and evaluates four defense configurations intervening at different stages of the persona-skill lifecycle, covering "online versus post-hoc intervention and active risk suppression versus passive backdoor protection". Privacy Sanitization rewrites user queries online to remove private cues while preserving task intent; Adversarial Obfuscation appends conflicting private attributes post-hoc to mislead distillation; Semantic-level Backdoor Injection implants semantic watermarks in both settings so unauthorized skill reuse can later be traced.
Design Implications¶
Context¶
Requirements¶
- The defender can only intervene on the trace side before distillation, transforming or augmenting the traces without modifying the distillation function, the agent, or the deployed skill-use interface
Constraints¶
- Active defenses primarily suppress surface-level communication cues while leaving deeper background and personality information exposed; backdoor-based protection degrades under persona-centric abstraction
Target Learners¶
- LLM-based agent systems and their developers
Target Learning Goals¶
- Protecting distilled persona skills against privacy leakage, impersonation, and unauthorized reuse
Claims¶
- Active defenses (Privacy Sanitization, Adversarial Obfuscation) reduce but do not eliminate persona-skill leakage, leaving personality and background exposed [+W]
- Passive backdoor defense (SBD) effectiveness is distillation-dependent, degrading under persona-centric abstraction [+W]
Related Strategies¶
Examples¶
-
Key Sources¶
- Yongli Xiang, Zhifang Zhang, Bojun Yang, Ziming Hong, Lei Feng, Miao Xu, Tongliang Liu. (2026). When Agents Learn to Be You: Benchmarking Privacy Leakage, Impersonation Risk, and Defenses in Persona Skills. arXiv. https://arxiv.org/abs/2608.03700
Page metadata
| Field | Value |
|---|---|
type |
strategy |
id |
persona-skill-defense-suite-strategies |
title |
Defense suite for persona skills: online sanitization, post-hoc adversarial obfuscation, and semantic backdoor watermarking |
description |
AntiSkillBench designs and evaluates four defense configurations intervening at different stages of the persona-skill lifecycle, covering "online versus post-hoc intervention and active risk suppression versus passive... |
status |
review |
generated |
by process:wiki-ingest on 2026-10-10 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — yongli-xiang-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.