mlsanityData to 7 Oct 2026

PrivacySkills: How Privacy Guidance Shapes Source Selection in LLM Agents

arXiv 2609.35937 ↗First posted 28 Sep 2026cs.CR cs.AI cs.HCcited by 0

On the map: LLM agents and tool use → Agent threats, guardrails and governance

Abstract (arXiv)

While prior work has documented privacy failures in LLM agents, it remains unclear how the presentation of privacy guidance influences their choice of information sources. We introduce PrivacySkills, a controlled framework for evaluating how agents choose among acquisition pathways that provide the same task-relevant value: consulting publicly available personal information, accessing confidential sources, or interacting with the user. The evaluation framework comprises 55 synthetic tasks spanning 11 categories of personal information, with 169 associated skills that describe the available acquisition pathways. We consider privacy guidance through system-level instructions, skill-level metadata labels, or both. Separately, we vary user availability and urgency framing. With users available and no privacy guidance, agents access confidential sources in 30% of valid runs on average across five open-weight models, despite sufficient alternatives. This rate increases to 45% when users are unavailable, whereas urgency framing has no detectable effect. System-level privacy instructions alone have limited effects on confidential access, while skill-level intrusiveness labels produce a modest reduction (24% on average), but combining the two roughly halves confidential access. Our findings motivate incorporating privacy annotations into skill specifications and evaluating their effectiveness alongside system-level instructions.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use. Builds on 25 of its 44 references that are in the map. First posted on arXiv 28 Sep 2026.

Builds on

25 of 44 references.

Closest ancestors · 12

Same direction, newest first

  1. 2026ToolMinimize: Auditing and Rewriting LLM Agent Tool Calls to Minimize Privacy Exposure25 Aug
  2. 2026ToolPrivacyBench: Benchmarking Purpose-Bound Privacy in Tool-Using LLM Agents26 Jun
  3. 2026PrivacyAlign: Contextual Privacy Alignment for LLM Agents19 Jun
  4. 2026When Lower Privileges Suffice: Investigating Over-Privileged Tool Selection in LLM Agents18 Jun
  5. 2026PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say29 May
  6. 2026Contextualized Privacy Defense for LLM Agents3 Mar
  7. 2026AgentLeak: A Benchmark for Internal-Channel Privacy Leakage in Multi-Agent LLM Systems12 Feb
  8. 2025Privacy in Action: Towards Realistic Privacy Mitigation and Evaluation for LLM-Powered Agents22 Sep
  9. 2025Searching for Privacy Risks in LLM Agents via Simulation14 Aug
  10. 2025Progent: Securing AI Agents with Privilege Control16 Apr
  11. 2025AgentDAM: Privacy Leakage Evaluation for Autonomous Web Agents12 Mar
  12. 2024AirGapAgent: Protecting Privacy-Conscious Conversational Agents8 May

Other references · 13

  1. 2023Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theorycited by 218
  2. 2024PrivacyLens: Evaluating Privacy Norm Awareness of Language Models in Actioncited by 161
  3. 2026SoK: Agentic Skills -- Beyond Tool Use in LLM Agentscited by 120
  4. 2025Securing AI Agents with Information-Flow Controlcited by 108
  5. 2026Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillationcited by 53
  6. 2024CI-Bench: Benchmarking Contextual Integrity of AI Assistants on Synthetic Datacited by 34
  7. 2025Tool Preferences in Agentic LLMs are Unreliablecited by 13
  8. 2025PrivWeb: Unobtrusive and Content-aware Privacy Protection For Web Agentscited by 8
  9. 2025BiasBusters: Uncovering and Mitigating Tool Selection Bias in Large Language Modelscited by 8
  10. 2025Operationalizing Data Minimization for Privacy-Preserving LLM Promptingcited by 8
  11. 2026How Many Tools Should an LLM Agent See? A Chance-Corrected Answercited by 7
  12. 2024What Affects the Stability of Tool Learning? An Empirical Study on the Robustness of Tool Learning Frameworkscited by 4
1 more
  1. 2026IDP-Bench: Benchmarking ability of LLMs to protect personal information in interdependent privacy contextscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Autonomy Reshapes How Personalization Affects Privacy Concerns and Trust in LLM Agents
  2. 2026It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs
  3. 2026ASLEval: Measuring Privacy Exposure Displacement in LLM Agent Sessions
  4. 2024Privacy Leakage Overshadowed by Views of AI: A Study on Human Oversight of Privacy in Language Model Agent
  5. 2025Beyond Jailbreaking: Auditing Contextual Privacy in LLM Agents
  6. 2026The Claws in Plain Sight: Unauthorized Context Disclosure through LLM Agent Tool Calls
  7. 2026Agents That Know Too Much: A Data-Centric Survey of Privacy in LLM Agents
  8. 2026PiSAs: Benchmarking Contextual Integrity in Multi-User Agentic Systems