Privacy-Preserving Text Sanitization for Distributed Agents Collaboration via Disentangled Representations
Addresses privacy leakage from distributional signatures in text exchanged between distributed agents, a problem for organizations using multi-agent collaboration.
DiSan is a privacy-preserving text sanitization framework for distributed multi-agent collaboration that disentangles text into a task-relevant role subspace and a source-identifying style subspace. It reduces PII exposure by 20× and stylometric attribution by over 70% while maintaining 83% answer faithfulness.
When distributed agents exchange text across organizational boundaries, privacy leakage arises not only from explicit identifiers but also from distributional signatures such as formatting conventions, vocabulary choices, and syntactic patterns. We propose DiSan(Disentangled Sanitization), a privacy-preserving sanitization framework and a built-in component of Intern-Shannon for multi-agent collaboration. DiSan uses a two-stream encoder to factorize text into a source-invariant role subspace that preserves task semantics and a source-identifying style subspace that remains local. Federated proto-type alignment and adversarial regularization enable joint training without centralizing raw text. Experiments show that identifier-level masking is insufficient: masking 19.2% of tokens reduces TF-IDF stylometric attribution by only 18.6%. By contrast, DiSan reduces answer-level PII exposure by 20 times while maintaining 83% answer faithfulness on a distributed multi-agent RAG benchmark, and lowers Enron stylometric attribution by 73.2% under TF-IDF and 70.6% under a neural probe.