It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs
cs.LG, cs.AI, cs.CR
Submitted: 2026-05-18
Updated: 2026-11-01
Comments: 29 pages, 16 figures
Code: https://github.com/sw-programmer/SelfCI
License: http://creativecommons.org/licenses/by/4.0/
The gist: Contextual Integrity (CI) defines privacy not merely as keeping information hidden, but as governing information flows according to the norms of a given context.
Terminology
Abstract
Contextual Integrity (CI) defines privacy not merely as keeping information hidden, but as governing information flows according to the norms of a given context. As large language models are increasingly deployed as personal agents handling sensitive workflows, adhering to CI becomes critical. However, even frontier models remain unreliable in making disclosure decisions, and existing mitigation strategies often degrade underlying task performance. To overcome this privacy-utility trade-off, we propose SELFCI, a complementary self-distillation framework that decouples information suppression from task resolution. SELFCI jointly optimizes two independent reverse KL divergences over distinct teacher distributions derived from feedback: one encourages preserving task-relevant information for utility, while the other enforces minimal and appropriate disclosure. This complementary formulation induces a Product-of-Experts (PoE) target, aligning the policy with the intersection of capability and privacy requirements. Empirical evaluations demonstrate that SELFCI, without relying on costly external supervision, consistently outperforms competitive baselines such as online reinforcement learning algorithms (e.g., GRPO). These trends further extend to out-of-domain settings involving agentic workflows and accumulated private context, suggesting that SELFCI provides a practical path toward CI alignment.
Sources
- Firewalls to Secure Dynamic LLM Agentic Networks
- CI-Bench: Benchmarking Contextual Integrity of AI Assistants on Synthetic Data
- Chain-of-Sanitized-Thoughts: Plugging PII Leakage in CoT of Large Reasoning Models
- Operationalizing Contextual Integrity in Privacy-Conscious Assistants
- The Llama 3 Herd of Models
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- Reinforcement Learning via Self-Distillation
- ThinkSafe: Self-Generated Safety Alignment for Reasoning Models
- Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security
- PrivacyBench: A Conversational Benchmark for Evaluating Privacy in Personalized AI
- Olmo 3
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- Self-Distillation Enables Continual Learning
- OpenAI GPT-5 System Card
- Learning by Distilling Context
- PrivacyReasoner: Can LLM Emulate a Human-like Privacy Mind?
- MPCI-Bench: A Benchmark for Multimodal Pairwise Contextual Integrity Evaluation of Language Model Agents
- Qwen2.5 Technical Report
- Qwen3 Technical Report
- Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis