DC-Area Anonymity, Privacy, and Security Seminar
Fall 2026 SeminarFriday, November 13th, 2026
Georgtown University
Hosts: Elissa Redmiles and Yoshi Kohno
| Program: | 1 p.m. – 5 p.m. |
| Room 150 | |
| Rafik B. Hariri Building | |
| McDonough School of Business | |
| Happy Hour: | 5 p.m. – 6 p.m. |
| Hilltop Tap Room | |
| Healey Family Student Center |
1:00 p.m. – 1:05 p.m.
Introductory Remarks
Paul Syverson (U.S. Naval Research Laboratory)
1:05 p.m. – 1:25 p.m.
Speaker: Christine Task (Knexus Research)
Title: DPDAP-Social: Differential Privacy for Social Platform Data Access
Abstract: Major public technology platforms, such as search engines and social media
networks, contain extensive information of public interest. Some platforms
are compelled to share information due to regulations, and others are
contemplating public-interest data sharing programs. Privacy risk is a
major barrier to more platform openness. Another key barrier to
transparency is the sheer size and complexity of data on these platforms.
We propose differentially private data access primitives (DPDAPs) as a
toolkit to a) reduce data complexity, b) standardize data outputs,
c) reduce privacy risks, and d) encourage data sharing from tech platforms.
DPDAPs can be used to safely distill insights from large platform data.
This talk will introduce the DPDAP conceptual framework and provide a
working demo of a DPDAP on social platform data. DPDAPs can be extended
into a host of data types and use cases.
"DPDAP-Social" operates on social platforms in particular (e.g., Reddit, Bluesky). It first ingests platform content using term co-occurrence matrices, scaling weights to normalize at the user level, summing across all users, and then adding noise to satisfy differential privacy. The summed matrices are then pruned, partitioned, and used to identify "story" topics that are active in each target community; post-processing transforms this into tabular data where it's easy to track the popularity and movement of stories amongst different communities across the platform.
1:25 p.m. – 1:45 p.m.
Speaker: Pierre Tholoniat (Work done at Columbia University, now at Google)
Title: Big Bird: Resilient Privacy Budgeting Across Untrusted Web Domains [paper]
Abstract: The W3C Attribution API is an emerging
standard for privacy-preserving advertising measurement. Its original
privacy architecture enforces individual differential privacy (IDP)
independently for each domain (e.g., an advertiser) issuing queries, using
the on-device accounting framework we proposed in Cookie Monster (SOSP '24).
In this talk, I will focus on our recent SOSP '26 paper, Big Bird. First, we show new soundness issues for realistic differentially private systems like Attribution, that exhibit cross-querier data adaptivity or enforce shared limits across queriers. The issue is not the on-device accounting model itself (device-epoch IDP) but treating each querying domain in isolation. Second, we propose Big Bird, a privacy-budget manager that makes global device-epoch IDP, enforced jointly across all domains, both sound and deployable for Attribution. Big Bird addresses the main obstacle to global enforcement in open multi-querier systems: denial-of-service depletion of a shared global budget by Sybil web domains. Both contributions are now part of the draft standard and have broader implications for DP systems in open, untrusted environments.
Finally, I will briefly discuss our work on architecting differential privacy into systems such as web browsers and datacenters, and share some lessons learned while building such practical DP systems.
1:45 p.m. – 2:15 p.m.
Coffee Break
2:15 p.m. – 2:35 p.m.
Speaker: Alexis Cao (Johns Hopkins University)
Title: Membership-Private Registration Based Encryption
Abstract: While encrypted communication protects message contents, the public-key lookup
required to initiate a conversation often reveals whether a phone number,
username, or email address is registered with the system. This creates a
membership oracle, allowing attackers to query large numbers of identities.
Large-scale enumeration attacks have been demonstrated against Signal,
WhatsApp, and Telegram. And consequences of such attacks can be substantial:
especially for activists, whistleblowers, or those in sensitive regions,
merely revealing the fact that they are registered with an anonymous service
may place them under suspicion. Yet current approaches to mitigating such
membership leakage remain fragile.
In this talk, we investigate the problem of privacy preserving key discovery. We present an approach to achieving membership privacy with registration-based encryption (RBE), where senders can encrypt using a succinct set of public parameters rather than performing a key lookup for the recipient. Critically, we demonstrate that our schemes are practical for real-world deployment. With one million registered users, our implementation achieves ciphertexts of 12.32 KB and an amortized registration time of 64.4 milliseconds. We further provide estimates for up to one billion users and show that ciphertexts are only 19.02 KB and the estimated amortized registration time is 1.72 seconds.
2:35 p.m. – 2:55 p.m.
Speaker: Ellis Fenske (U.S. Naval Academy)
Title: Inferring Tor Guard Sets by Observing Middle Relays
Abstract: This talk will discuss a new approach to performing guard
discovery and user linkability attacks on Tor Browser users from the
perspective of honest-but-curious exit relays. The inference technique is
based on repeatedly observing the middle relay selected for a circuit,
which could allow the adversary to learn that client's guard set and link
sessions together. We show this can be made practical by combining a
variety of independent mechanisms to amplify the rate of adversarial
observations and assess its effectiveness with simulations and experiments
on the Tor network.
2:55 p.m. – 3:15 p.m.
Speaker: James Parker (Ossa)
Title: Ossa: Solving the Dueling Admin Problem for Secure Decentralized Applications [code]
Abstract: Decentralized collaborative applications have the potential to offer improved
resilience, privacy, and security over traditional cloud applications.
Unfortunately, the Dueling Admin problem is an open problem that prevents
decentralized applications from gaining widespread adoption. The Dueling Admin
problem arises when two administrators mutually revoke access from one
another. In this scenario, it is unclear who the winner should be and who
should retain their permissions.
In this work, we solve the Dueling Admin problem by employing a hybrid consistency model. To accomplish this, we define a new asynchronous Byzantine fault tolerant consensus protocol designed for this setting. We build upon this breakthrough to build the Ossa distributed system. Ossa enables developers to easily build decentralized collaborative applications that are peer-to-peer, end-to-end encrypted, Byzantine secure, and protected by role-based access control. We evaluate Ossa with three case study applications and benchmark its consensus protocol against prior work.
3:15 p.m. – 3:45 p.m.
Coffee Break
3:45 p.m. – 4:05 p.m.
Speaker: Wentao Guo (University of Maryland)
Title: An exploratory user study of disclosure avoidance tools for scientific microdata [paper]
Abstract: Social, behavioral, and health scientists are often required or
encouraged to de-identify and publish data to support open science. Prior
work shows that scientists, lacking support, avoid systematic disclosure
avoidance methods that could help them achieve their own privacy goals more
rigorously. Better tools could help but must be tailored to scientists'
real-world needs in order to have an impact. To understand these design
needs, we conducted an exploratory user study in which experienced
researchers used one of two disclosure avoidance tools on a real research
dataset. In my talk, I will highlight two main findings. First, a high
degree of skepticism about disclosure risk in the original data led
participants to perceive these tools as extreme. Second, these tools
contain promising design patterns for helping scientists understand impacts
on data utility, but they largely fail to improve scientists' understanding
of privacy. Our findings indicate a need for better approaches to evaluate
disclosure risk, and we provide a foundation for tailoring disclosure
avoidance tools to the needs of researchers participating in open science.
4:05 p.m. – 4:25 p.m.
Speaker: Rahel Fainchtein (Johns Hopkins University Applied Physics Laboratory)
Title: An Overview and Proposed Architecture for Digital Emblems
Abstract: As noted in their charter, the Digital Emblems (diem) IETF working group
broadly defines Digital Emblems as symbols denoting a status or
requirement(s) as defined or identified by specific laws, agreements or
standards, that is presented via digital communication channels.
In this talk, we discuss what this means in practice, briefly covering
potential use cases and noting the resulting requirements these emblems
would have respectively.
Given the current scope limitations requiring that Digital Emblems be
implemented in DNS, we then go over the questions surrounding how
emblems should be structured currently being discussed in the working group
and pose our answers in the form of a brief walkthrough of our proposed
architecture and the reasoning underlying
our structural choices.
4:25 p.m. – 4:30 p.m.
Closing Remarks
Paul Syverson (U.S. Naval Research Laboratory)
4:30 p.m. – 5:00 p.m.
Discussions and Networking
5:00 p.m. – 6:00 p.m.
Happy Hour: Hilltop Tap Room in the Healey Family Student Center
Driving: Parking is limited on campus, but there is parking nearby at the Southwest Garage. It costs $25/day.
Metro: The nearest Metro stops are Rosslyn (on the Blue, Orange, and Silver lines) and Dupont Circle (on the Red line). These are 1.3 miles and 2.0 miles from the Rafik B. Hariri Building, respectively. Georgetown University runs a regular shuttle service from each of these stops to campus: GUTS. The Lombardi Circle and McDonough McDonough Bus Turnaround stops are 0.2 and 0.3 miles to the Rafik Hariri Building, respectively. The shuttle is free, and anyone coming to the university can take the shuttle, no ID or documentation is required. Note that some shuttles have a sign that says MedStar Lombardi, and a hospital ID would be required to ride those shuttles. Wi-Fi: You can connect via eduroam if you have an eduroam account. Otherwise, you can use Guestnet for basic web access.