CuraSec

tag: Llm-Privacy · 4 items

Learn Attribute Inference Attack on Edge-Cloud Split Multimodal LLM Features
2026-10-26 · arXiv cs.CR · source ↗ #llm-privacy#split-inference#attribute-inference
  • Engineer — Learn: If you’re designing or auditing a split-inference LLM pipeline, this research challenges the assumption that keeping raw images on-device is sufficient privacy protection — intermediate features still leak age, gender, and race. No patch exists; the takeaway is an architectural one: evaluate representation-level privacy, not just response-level filtering, before deploying such pipelines.
  • SOC/IR — Skip
  • Leader — Learn: Relevant for AI governance: if your organization deploys split-inference LLMs that process biometric attributes, this research demonstrates that edge-local raw data does not prevent cloud-side inference of protected characteristics — a GDPR/HIPAA consideration worth flagging to your AI/privacy counsel when reviewing new deployments.
Learn Veil-then-Reveal: PII Pseudo-Anonymization for Zero-Trust LLM Use
2026-10-19 · arXiv cs.CR · source ↗ #llm-privacy#pii-protection#applied-research
  • Engineer — Learn: Novel framework for replacing PII with realistic same-type surrogates before prompts leave the device, preserving linguistic context while keeping plaintext off remote providers. Worth evaluating if building systems that feed sensitive documents into external LLM APIs.
  • SOC/IR — Skip
  • Leader — Learn: This research surfaces a practical gap: organizations using external LLM APIs may inadvertently expose PII through prompt logging and retention — useful framing for an AI data-governance policy before the next audit cycle.
Learn archived FHE Inference System Odin Enables Privacy-Preserving Llama 3 Deployment
2026-09-14 · arXiv cs.CR · source ↗ #fhe#llm-privacy#cryptography
  • Engineer — Learn: Odin’s co-designed ciphertext packing approach for FHE LLM inference is a technique to watch if your team is evaluating privacy-preserving cloud inference architectures; no running systems need changes today.
  • SOC/IR — Skip
  • Leader — Learn: Early academic work demonstrating that cloud LLM inference without decrypting user prompts is feasible at scale; useful context when briefing on AI privacy posture or evaluating vendor data-handling commitments.
Learn archived P2Skill: PII-aware prompt routing for cloud-local LLM inference
2026-08-17 · arXiv cs.CR · source ↗ #llm-privacy#pii#ai-inference
  • Engineer — Learn: Novel prompt-based technique for splitting LLM inference between local and cloud without leaking PII; worth tracking if you’re building hybrid AI pipelines, but no patch or config action required today.
  • SOC/IR — Skip
  • Leader — Learn: Relevant background for leaders defining AI data governance policies around cloud LLM usage, but no immediate risk register or vendor action required.