PEPR '25
as the organisers put it
The 2025 USENIX Conference on Privacy Engineering Practice and Respect (PEPR '25) will take place on June 9–10, 2025. PEPR is focused on designing and building products and systems with privacy and respect for their users and the societies in which they operate. Our goal is to improve the state of the art and practice of building for privacy and respect and to foster a deeply knowledgeable community of both privacy practitioners and researchers who collaborate towards that goal.
who it's for
privacy engineers, machine learning engineers, data engineers, product managers focused on privacy, privacy researchers, security engineers, compliance managers
topics
differential privacy production deployments; differential privacy visualizations for stakeholders; genomic data privacy metrics; federated learning across healthcare systems; secure multi-party computation analytics; large-scale deidentification pipelines; consent and preference management platforms; observability-safe telemetry design; privacy for LLMs and synthetic data; LLM-assisted incident clustering and triage; notice-and-choice threat modeling; graph data reidentification risk
| title | author | length |
|---|---|---|
| Privacy Paradigms for Law Enforcement Response | Lukas Bundonis, Ben Ballard | 20m |
Claims that corporate law enforcement response systems are opaque and attractive crown jewels for nation-state attackers, citing the 2024 Salt Typhoon… | ||
| Remediating Systemic Privacy Incidents | Sam Havron | 14m |
Built a system that clusters privacy incidents by root-cause using a weighted-graph plus community-detection and LLM scoring, runs scheduled llama inference… | ||
| Observable...Yet Still Private? An Offensive Privacy Perspective on Observability | Cat Easdon, Patrick Berchtold | 17m |
A prototype statistics-only API aims to give analysts trends without raw RUM data while avoiding differential privacy; reconstruction attacks use… | ||
| Using Privacy Infrastructure to Kickstart AI Governance: NIST AI Risk Management Case... | Katharina Koerner, Nandita Rao Narla | 12m |
Argues that existing privacy tooling can bootstrap NIST AI RMF adoption by reusing concrete artifacts: a startup leverages a data catalog fed model inputs and… | ||
| UsersFirst: A User-Centric Threat Modeling Framework for Privacy Notice and Choice | Norman Sadeh, Lorrie Faith Cranor | 16m |
UsersFirst is a threat-modeling framework that breaks notice-and-choice failures into four actionable categories—discovery and use, comprehension problems… | ||
| Enterprise-Scale Privacy for AI: How Canva Scaled Customer Control of Data for AI... | Phillip Ward | 15m |
Canva describes an end-to-end consent platform that separates content and usage toggles, runs behind the data access layer, and enforces ~1,000 jurisdiction… | ||
| My $5MM Differential Privacy Visualizations | Marc-Antoine Paré | 11m |
Project claims that visualizations unlock $5MM in funding and convince regulators and data scientists by showing real outputs and tradeoffs rather than… | ||
| Establishing Privacy Metrics for Genomic Data Analysis | Curtis Mitchell | 14m |
Multi-agency project builds privacy-preserving federated learning pipelines on soybean genomic data as a human-data stand-in, using PyTorch CNNs, Opacus DPSGD… | ||
| Practical Considerations for Differential Privacy | Alex Kulesza | 19m |
Seven years of building DP systems at Google produce concrete lessons: production libraries used by hundreds of teams, a recurring inbounding sensitivity… | ||
| Unlocking Cross-Organizational Insights: Practical MPC for Cloud-Based Data Analytics | Daniele Romanini | 16m |
Built a cloud-native MPC pipeline using Carbyne Stack and MP-SPDZ, adding support for a semi-honest model and bits/dits primitives, plus a sorting+bucketing… | ||
| Building an End-to-End De-Identification Pipeline for Advertising Activity Data at... | Saikrishna Badrinarayanan, Chris Harris | 16m |
LinkedIn builds an end-to-end deidentification pipeline that processes billions of events a day using both streaming (Samza) and batch paths with QoS tiers to… | ||
| Network Structure and Privacy: The Re-Identification Risk in Graph Data | Daniele Romanini, Sune Lehmann, Mikko Kivelä | 10m |
Finds that neighborhood structure alone re-identifies nodes: authors build uniqueness maps to visualize how many nodes have structurally unique neighborhoods… | ||
| Data Classification at Scale: Taming the Hydra | Daniel Gagne | 18m |
Describes a bytes-up approach that builds a taxonomy with semantic types and actor facets, scans assets via enumerator → scheduler → scanner, extracts feature… | ||
| Harnessing LLMs for Scalable Data Minimization | Charles de Bourcy | 15m |
Claims that foundation LLMs outperform regex for messy column formats by labeling table/column/examples into categories like name phone email other, and that… | ||
| Demystifying the Android Telehealth Ecosystem | Primal Wijesekera | 20m |
Analyzes 408 Android telehealth apps from 36 countries using an OS-level interception tool that achieves an 88% interception rate, finds widespread… | ||
| Safetypedia: Crowdsourcing Privacy Inspections | Lisa LeVasseur, Bryce Simpson | 13m |
Internet Safety Labs runs the SafetyPedia pilot (since Oct 2024) to crowdsource mobile app privacy inspections while also developing automation, aiming for… | ||
| Verifying Humanness: Personhood Credentials for the Digital Identity Crisis | Tanusree Sharma | 19m |
Compares deployment trade-offs for personhood credentials using a policy review of 18 regulatory frameworks, lab and interview studies with US and EU… | ||
| Career Advice for Privacy Engineers: From Resume to Interview to Finding the Next Job | Jason A. Novak | 17m |
Argues that privacy engineering jobs split into four concrete quadrants (programmatic vs engineering crossed with product-differentiator vs compliance) and… | ||
| Privacy Engineers on the Front Line: Bridging Technical and Managerial Skills | Ramazan Yener | 12m |
Study surveys 28 privacy engineers and interviews 18 practitioners, experience 2–25 years, across tech, healthcare, retail and firms up to 25,000 employees… | ||
| Purpose Limitation with Policy Zones: What We Tried That Didn't Work | Diana Marsala | 11m |
Meta describes why three early Policy Zones designs fail at scale: dynamic propagation breaks on cycles and asynchronous jobs so policy changes ripple… | ||
| Building Privacy Products: Field Notes | Miguel Guevara | 13m |
Internal-facing data-minimization effort uses an email wireframe and simple heuristics (files older than two years) to test demand, then adds last-access… | ||
| Panel: How Privacy Engineers Can Shape the Coming Wave of AI Governance | Zachary Kilhoffer, Hoang Bao, Masooda Bashir, Debra J Farber, Sarah Lewis Cortes, Akhilesh Srivastava | 33m |
Panel argues privacy engineers should operationalize AI governance now: use data catalog tagging for model governance and extend cookie-banner disclosure… | ||
| Beyond RAG: Building Reliable AI Systems for Privacy Assessments | Emily Choi-Greene | 14m |
Ten minutes of practical engineering: replaces plain RAG with tool use / function calling to call APIs and avoid hallucinations, a claim-evaluation loop that… | ||
| When Privacy Guarantees Meet Pre-Trained LLMs: A Case Study in Synthetic Data | Yash Maurya, Aman Priyanshu | 10m |
Analysis finds a differentially private synthetic financial dataset on Hugging Face with about 1000 synthetic records derived from ~14,000 SEC filings… | ||
| Breaking Barriers, Not Privacy: Real-World Split Learning across Healthcare Systems | Sravan Kumar Elineni, Sean Muir, Dave Carlson, Himali Saitwal, David E Taylor, Chit win, Jayme Welty, Adam Wong, Jordon Everson, Keith Salzman, Serafina Versaggi, Lindsay Cushing, Savannah Mueller | 17m |
Three independent HIEs jointly train a deep model using split learning (no raw data leaves sites) and report successful training on five comorbidities… | ||
| OneShield Privacy Guard: Deployable Privacy Solutions for LLMs | Shubhi Asthana | 14m |
OneShield Privacy Guard is a deployable framework integrated with IBM Krenite that enforces real-time, context-aware privacy across input, inference and… | ||
| Using GenAI to Accelerate Privacy Implementations | Rituraj Kirti | 12m |
Meta shows a concrete pipeline that uses Code Llama–based classifiers and prompt engineering to auto-annotate, review, prevent regressions in… | ||
| From Existential to Existing Risks of Generative AI: A Taxonomy of Who Is at Risk,... | Megan Li, Wendy Bickersteth | 16m |
Presents a coordinated taxonomy built from 499 incidents that maps 14 sociotechnical failure modes to 41 specific harms and affected stakeholders, finding most… | ||
| nothing matches | ||