PEPR '26
as the organisers put it
The 2026 USENIX Conference on Privacy Engineering Practice and Respect (PEPR '26) will take place on June 1–2, 2026. 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, software architects, data protection officers, platform engineers, product managers, compliance engineers
topics
differential privacy pipelines for chatbots; differentially private synthetic data; private fine-tuning of LLMs; production multi-party computation; data retention service architecture (Canva); scalable data lineage graphs; data portability via HTTP + OAuth and JSON schema; age verification and identity wallet futures; LLM-assisted privacy impact assessments; privacy-aware model interfaces and provenance; privacy threat modeling at scale; privacy engineering for local government adoption
| title | author | length |
|---|---|---|
| Envisioning and Mitigating Privacy Risks for Consumer-Facing AI Product Concepts... | Hao-Ping (Hank) Lee | 15m |
Privy is a GenAI human-AI tool that helps non-experts draft privacy impact assessments by generating use cases, surfacing risks from an AI privacy taxonomy… | ||
| From Legalese to Logic: Translating Privacy Requirements into Verifiable Controls via... | Rituraj Kirti, Inchara Shivalingaiah | 19m |
An LLM-powered pipeline that translates textual privacy requirements into implementable tasks uses a deterministic runner plus an agent network and Python… | ||
| A DP Framework for Gaining Insights into AI Chatbot Use | Mikhail Pravilov, Edith Cohen, Vadym Doroshenko, Badih Ghazi, Charlie Harrison, Peter Kairouz, Pritish Kamath, Alexander Knop, Ravi Kumar, Ethan Leeman, Daogao Liu, Pasin Manurangsi, Adam Sealfon, Da Yu, Chiyuan Zhang | 13m |
Urania converts millions of raw chatbot logs into DP-safe summaries and trends, claiming end-to-end ε,δ DP guarantees by privatizing cluster centers and… | ||
| CA-CI: A Normative Framework for Evaluating Privacy and Dignity in AI Governance | Florian Schaub | 17m |
CA-CI claims two concrete moves: add purpose as a sixth parameter to Contextual Integrity and use Nussbaum’s 10 capabilities as minimum dignity thresholds. The… | ||
| Vision: Human-as-the-Unit Privacy Management with AI Agents | Eryue Xu, Tianshi Li | 07m |
Interviews with 12 users surface nine cross-context privacy challenges spanning apps, time, and relationships, and researchers generate nine agent concepts to… | ||
| Enforcement of Data Protection Laws in Africa: Implications for Privacy Engineers | Mubarak Raji | 12m |
Forty-four of 55 African Union countries have data protection laws as of Dec 2025, and a 20-country analysis finds all laws cover public and private actors… | ||
| Dismantling the Barriers to Personal Data Portability | Arthur Borem, Lisa Dusseault | 15m |
A survey of ~500 platforms finds <3% support server-to-server transfers, and most portability still uses manual export ZIPs that can be dozens of gigabytes… | ||
| Why Consent Fails in Practice: Lessons from Web Measurement Studies | Muhammad Abu Bakar Aziz, Christo Wilson | 16m |
A custom crawler instruments 7,000 California publishers to click banners, send a Global Privacy Control header, log IAB/GPP/USP APIs and HTTP requests, and… | ||
| User (Non-)Compliance with Age Verification: Evidence from a Deceptive Web Experiment | Lorrie Faith Cranor | 16m |
A deceptive web experiment with 1,635 participants tests five verification methods via an AgeGuardian iframe prototype and simulated camera and ID flows… | ||
| Turning Privacy Risk Assessment Into 20 Questions for Developers | Qiyu Li, Haojian Jin | 15m |
Prototype PrivacyAkinator converts privacy design into LLM-generated multiple-choice flows that prompt roughly 20 closed-ended questions, using a verb-based… | ||
| Panel: The AI Architecture Debt—Refactoring Infrastructure for Sustainable Privacy | Dylan Gilbert, Sri Pravallika Maddipati, Nathalie Baracaldo, Gary Young | 31m |
Panel argues that AI systems accrue concrete architecture debt when models and RAG pipelines are wired without data provenance logging or the ability to honor… | ||
| How Canva Built Simple, Auditable, and Maintainable Data Retention | Phillip Ward | 16m |
Canva re-architects deletion with a central deletion service using a star topology that replaces brittle event chains, generates per-service wiring from a… | ||
| Architecting Scalable Data Lineage Graph for Privacy Compliance and Agentic Analysis | Maharshi Jha, Aygun Aydin | 15m |
Meta builds an in‑memory lineage graph that ingests from 15+ lineage providers and processes billions of edges daily across web, warehouse, and AI systems. The… | ||
| Responding to Regulations Confidently and Efficiently | Marek Matejka | 13m |
Framework claims to turn setting-related regulatory requirements from weeks or months into code-complete in one working day by using a declarative, one-file… | ||
| Designing for Civic Trust: An Infrastructure to Help Long Beach Residents Manage Their... | Omar Moncayo, Norman Sadeh, Gwen Shaffer | 19m |
A City of Long Beach deployment pairs CMU’s IoT Privacy Infrastructure with on-street data privacy labels (24 tested; March 2024) and QR-linked entries to a… | ||
| Provenance Without Surveillance: Privacy Engineering for AI Content Transparency | Sai Prashanth Chandramouli, Sankalp Jain, Gayathri Ravi | 20m |
Argues that C2PA-style provenance is dual-use and offers a concrete privacy playbook: a deployed pipeline that collapses action logs, uses org-level signing… | ||
| Toward Provably Private Insights into AI Use | Rakshita Tandon, Albert Cheu, Artem Lagzdin, Brett McLarnon, Daniel Ramage, Katharine Daly, Marco Gruteser, Peter Kairouz, Stanislav Chiknavaryan, Timon Van Overveldt, Zoe Gong | 14m |
A system called Provably Private Insights combines trusted execution environments with a “data expert” LLM and differential privacy to ensure raw user data… | ||
| DPSynth: From Research to Production—Engineering Differentially Private Synthetic... | Mikhail Pravilov, Ryan McKenna, Peter Kairouz, Alexander Knop, Vadym Doroshenko, Eva Bertels | 13m |
DPSynth is a marginal-based DP library that turns research into a production system: it measures low-dimensional marginals with epsilon-delta guarantees, fits… | ||
| Generating High-Quality Tabular Synthetic Data at Scale | Shripad Gade, Samuel Maddock, Graham Cormode, Will Bullock | 16m |
Introduces GEM+ as a generator‑network replacement for probabilistic graphical models that scales synthetic tabular generation to hundreds and up to ~1,000… | ||
| Profile-Then-Simulate: Can LLMs Faithfully Generate Differentially Private Synthetic... | Nassima Bouzid, Dehao Yuan, Nam H. Nguyen, Mayana Pereira | 13m |
Profile-then-Simulate takes 2,000 users and 24M transactions, distills 1,600 users into DP-protected personas (all epsilon spent there) and seeds PersonaLedger… | ||
| Private Tuning of LLMs in Practice: From VaultGemma to Custom Fine-Tuning | Amer Sinha, Ryan McKenna, Galen Andrew, Borja Balle, Vadym Doroshenko, Arun Ganesh, Weiwei Kong, Alex Kurakin, Brendan McMahan | 17m |
Claim: Differential Privacy is practical at production scale using JAX Privacy to train and fine-tune Gemma-family models including VaultGemma. Implementation… | ||
| Private AI: Building Trust Through Verifiable Computation | Mingshen Sun, Mateus Guzzo | 19m |
Claims that user data never becomes visible to the provider while still running LLM workloads on cloud GPUs and that auditors can verify that claim remotely… | ||
| Production Multi-Party Computation via the Distributed Aggregation Protocol | Tim Geoghegan, J.C. Jones | 15m |
A production deployment called Divvi Up runs the IETF Distributed Aggregation Protocol to process billions of secret-shared telemetry reports over 3+ years… | ||
| Privacy in Theory, Bugs in Practice: Grey-Box Testing for Differential Privacy Libraries | Jack Fitzsimons, Tudor Cebere, David Erb, Damien Desfontaines, Aurélien Bellet | 17m |
An open-source framework called DP Recorder wraps mechanisms for unit tests and uses a record-and-replay approach to run a program on a dataset and its… | ||
| The Emperor's New Embeddings: Obfuscating ML Inputs Doesn't Provide Privacy | Jack Fitzsimons, Daniel Simmons-Marengo, Tudor Cebere, Damien Desfontaines | 21m |
Claims that token or embedding obfuscation gives inference privacy fail: applying the data processing inequality shows transformed inputs cannot hide more… | ||
| Surfacing Hidden Privacy Risks in Code: Lessons from LLM and Retrieval Assisted Detection | Bushra Aloraini, Jimmy Haslam | 18m |
A detection pipeline evaluates 183,000+ methods across nine repositories using GPT-4 with either fixed few-shot prompts or a retrieval-augmented generation… | ||
| V.O.I.C.E.: A Data-Driven Risk Taxonomy for Synthetic Voice Generation | Tanusree Sharma, Lili Dudas | 18m |
V.O.I.C.E. is a data-driven taxonomy built from 569 incidents, 2,221 Reddit discussions and 1,069 direct reports that classifies six risk categories and maps… | ||
| The Disposable Identity: Eliminating Non-Human Identity Risk in Federal Healthcare... | Tristan McKinnon | 17m |
Production deployment of an Identity-Per-Transaction pipeline on AWS replaces 90‑day service accounts with unique, cryptographically scoped ephemeral… | ||
| Shadow Data in Tool Calls: The Privacy Leak Hiding in Plain Sight | Shabista Shabista, Ravi Gupta, Mayank Kumar Raunak | 06m |
Local agents leak sensitive inputs inside external tool requests: prototype analysis of a smart‑home agent finds a 72% leak rate across sampled calls, with… | ||
| Adopting AI in Local Government with Privacy and Equity in Mind: A Case Study of the... | Shufan Chai, Vinal Dalcy Dsouza, Rebecca Williams Earle, Rasika Bhalerao, Jessica Staddon | 05m |
Partnership with the City of Oakland pilots Microsoft Copilot with an AI task force of 45 employees, collecting ten 45–60 minute interviews, three surveys and… | ||
| Mapping the Privacy Workforce in the AI Era | Ramazan Yener, Muhammad Hassan, Masooda Bashir | 15m |
Researchers scrape about 2,000 postings, clean to 1,143 job postings, and apply NLP and BERT topic modeling to map titles, skills, pay, and AI ties. Findings… | ||
| Privacy Review for Non-Maniacs | Katriel Cohn-Gordon, Ben Fonarov | 16m |
Meta reworks privacy review from manual checklists into upfront requirements plus automation, claiming ~500 privacy requirements and roughly half now enforced… | ||
| Training Developers' Privacy Awareness with Enforcement Cases | Shao-Yu Chu, Xu Wang, Haojian Jin | 14m |
Ace Privacy Detective turns FTC complaints (for example a 29‑page Primom complaint) into playable levels where investigators collect real evidence, assign… | ||
| Scaling Privacy Threat Modeling: From Architects to Developers | Nitish M. Uplavikar | 18m |
An open-source set of 12 privacy rules implements LINDDUN-derived checks as extensions to Threagile, taking a YAML architecture input and running a rule engine… | ||
| nothing matches | ||