Understanding Privacy Risks in Code Models Through Training Dynamics: A Causal Approach
This work addresses privacy concerns for developers and users of code models by providing causal evidence that leakage risks are type-dependent, offering guidance for type-aware defenses, though it is incremental as it builds on prior studies of PII in models.
The paper tackled the problem of privacy risks in code models by investigating how different types of personally identifiable information (PII) vary in their likelihood of being learned and leaked, finding that leakage risks differ substantially across PII types, with easy-to-learn instances like IP addresses showing higher leakage and harder types like keys and passwords leaking less frequently.
Large language models for code (LLM4Code) have greatly improved developer productivity but also raise privacy concerns due to their reliance on open-source repositories containing abundant personally identifiable information (PII). Prior work shows that commercial models can reproduce sensitive PII, yet existing studies largely treat PII as a single category and overlook the heterogeneous risks among different types. We investigate whether distinct PII types vary in their likelihood of being learned and leaked by LLM4Code, and whether this relationship is causal. Our methodology includes building a dataset with diverse PII types, fine-tuning representative models of different scales, computing training dynamics on real PII data, and formulating a structural causal model to estimate the causal effect of learnability on leakage. Results show that leakage risks differ substantially across PII types and correlate with their training dynamics: easy-to-learn instances such as IP addresses exhibit higher leakage, while harder types such as keys and passwords leak less frequently. Ambiguous types show mixed behaviors. This work provides the first causal evidence that leakage risks are type-dependent and offers guidance for developing type-aware and learnability-aware defenses for LLM4Code.