SOUPS | Kami Vaniea

Symposium on Usable Security and Privacy - Day 3

Wed, 26 Aug 2026 03:01:43 -0400

Welcome to the happening in Hannover, Germany and being hosted by CISPA. The Symposium is a community that studies how people use security and privacy technologies.

Blog posts:

Session: Usable Security Mechanisms & Emerging Threats

Presenter: Yorick Last, Paderborn University

You can sign digital documents via electronic or digital signatures. Handwritten signatures work well but provide no guarantees of the authenticity or the integrity. Often because they are typed, or could be easily copied.

One way to make this easier and better for users is Digital Identity Wallets applications that collect various identity attributes, presented to the user in a card-based interface. These are decentralized user-centric means of identiy management. EU is rolling out a Digital Identity Wallet next year.

The project proposes such a prototype wallet: IdentitySign. Free and open-source web applicatoin, uses the Yivi identity wallet and lends cryptographic support from .

Had a user study including an unmoderated usability test with 99 participants recruited accross the EU.

IdentitySign was generally found to be useable. But some users did not understand why such a signature was needed over just signing their name with a (digital) pen. Only 20% could detect a valid signature.

So the researchers tried to address the issue of misplaced trust. Also added some friction to the UI for verification to slow the user down.

Ran the study again with 74 participants. Again from across the EU and an unmoderated remote user study. The scenarios included 1 signing and 2 verification tasks.

Usability was still considered to be good. 82% could now correctly find broken signatures. 95% correctly identified a signature as correct.

A key lesson was that the UI needs to balance being helpful without seeming to imply trustworthyness. It is important that users be able to judge if a signature is correct themselves.

Presenter: Xueyang Wang

The visual-physical discrimination vulnerability - the point at which humans cannot differentiate between real and faked. This is also known as the .

This research project looks at ways that an attacker can inject virtual objects onto the user’s vision.

Endogenous injection
- invisible duplicates of existing objects in the user vision
- users have trouble detecting these
Exogenous Injection
Type-deceptive overlay
- Modifying a part of the user vision. For example adding flowers to the top of a trash can so that it looks like a flower pot. The user can still see it but cannot figure out where the trash can is.
Attribute-overlay
- Adding elements like putting logos on a white t-shirt can impact user decisions. Even when a user knows that the image was fake and had been added. This is well in alignment with marketing research.

Attacks with the highest behavioral success rates received the highest plausibility, lowest workload, and highest usability ratings. In other words, if people thought the fake thing was real, they were happy to engage with it and were happy about it.

Research outcome: even if the user detects the fake, it still impacts their decision making and physical behavior.

Presenter: Katarina Galanska, Masaryk University

Interested in how recommendations were acted on. Security recommendations tend to be high-level and general so many possible actions could be taken based on them.

Suggestions for how to address a security vulnerability were written by humans and various of AI. Did a survey-based study asking users about the following three aspects of each suggestion.

Clarity
Actonability
Confidence

Lightning talks

Florin Martius, Registered Reports at SOUPS: Call for Discussion
- Registered reports: comprehensive research plan is submitted to the reviewers rather than the final research paper. The idea is to not waste research time, and create more high-quality research.
- Challenges exist though, it now shifts reviewing effort.
- Kami’s opinion: Contextualization of results is important in a final publication. Because it is not just about the methods. It is also about the interpretation of the results which cannot be done up-front.
Simon Althaus, PIONEER - A PrIvacy companion for mOtivatioN and knowlEdge transfER
- Rough motivation is to make it easier for users to handle privacy
- App involves privacy and lessons and motivations
- Can compare apps to each other to make decisions
Christina Detsika and Julia Justen, Age Groups for Security and Privacy Studies with Children
- Deciding what age ranges of children to do research with is a bit complex because developmental stages can drastically impact cognitive understanding. The work aims to determine suitable age ranges for research in the USEC space as well as how to best talk about the population choices in research papers.
Camille Cobb, How do we encourage students (and ourselves) to take meaningful action against dystopian technologies?
- Advocating action-oriented tech ethics education
- Current ethics materials focus on how to make judgments. But not much about how to take action after the judgment is made. Especially when the action is professionally risky or it is unclear how to do.
Nicole Dircksen - What should security security researchers do in an ethical situation.
- Research project is aimed at understanding where researchers agree and disagree on what is ethical in the field. Aim is to better understand our current understanding to advise new rules.
- For example: researchers find a vulnerability in a medical device where the company is now out of business. Should they disclose so that hospitals know and can remove or better protect such machines BUT also inadvertently tell attackers about the issue. Or should they stay silent and hope that attackers do not also find the vulnerability.

Session: Scams, Fraud & Misinformation

Presenter: Smirity Kaushik, The George Washington University

$3 billion is lost to social media fraud in 2024 ( ) and short form videos are popular with the general population. This also leads to a whole ecosystem of influencers.

RQ’s

What types of scams?
How do users report scams?
How do TikTok users identify these steps?

The researchers manually searched user-reported videos for scams. Found 150 videos (after reviewing them).

Types of scams:

Financial
Influencer targeting
- asking influencers to do things like buy a product and review it, but the influencer has to buy first before getting compensated.
- cloneing an influencer and then asking for things. Causes the influencers to have to post about such scams
- Deepfake of an influencer to promote a different product than the influencer initially promoted.
Shopping
- Fake or poor-quality products
- Goods never delivered
Account
Romance
Miscellaneous

How do scammers entice people?

Lots of use of direct message and “for your page”
Scammers used “need and greed” or the “distraction” principles to lure victims

How were scams identified?

Found the same as the FTC common approaches like request to pay right away
Also found that scammer communication patterns, account discrepancy, unauthorized transactions, product and price mismatch

Kami’s view: this is reminding me of our paper on , I think that we need more research that looks at what people say after falling for a scam organically, that is what they write immediately after they fall for the scam, as opposed to what they say to a researcher which may be biased by the presence of the researcher.

[Source-Level Disengagement: A Usable Security Defense Against Misinformation](Source-Level Disengagement: A Usable Security Defense Against Misinformation)

Presenter: Bogdan Carbunar, Florida International University

Misinformation is a serious issue and even if users are aware it can still influence thinking as shown by the earlier paper today on MR and changing what users see.

This research targets the issue that the user is seeing repeat misinformation from a person they are following and are not unfollowing.

Three interventions

Inline Warning
- Warning below main post
Action Controls
- Warning followed by buttons like block and filter
Blur Overlay
- post cannot be seen without clicking
Control
- no intervention, just the post content

29 participants in a within-subjects study. Processed 4 timelines each of which had 12 posts (3 with misinformation).

The control condition had only one user do something about the misinformation. Bu tfar more with the other options, particularly the action controls option where controls were made readily available.

Users sometimes chose the blur or filter options over other actions like unfollowing because unfollowing leads to sending social signals and breaks a relationship.

B luring still caused participants to view the content anyway, and then took an action.

Kami question: I wonder if this finding would persist in the real world. Earlier comment is about filtering and blurring to avoid content without breaking a relationship. It may be that in the real world users get to know sources and therefore do not open the content because they know its likely shape already.

Design Implications:

It should be easy to act when misinformation is encountered, not just warned, users should have a clear path to act
Preserve relationships should be possible. Users should have silent blocking ability where the followed person does not know their content has been hidden. Users are more likely to take this option
Design for limits - platforms also need to take actions themselves. Users get overwhelmed, so the platform itself needs to reduce user burden.

Presenter: Nathan Malkin, New Jersey Institute of Technology on behalf of Zane Xu

Survey research is starting to have problems with LLMs giving answers instead of real users. This is super bad for science. Attention check questions like: *“please answer `strongly agree’ to this question” are irronically easier for LLM than humans are are more likely to be correctly answered. One option is questions that LLMs answer poorly, but well those are vanishing. We could also just ask participants “please do not use AI”, but well that doesn’t really work.

This research

Can we detect LLM
How often
What can we do

So a survey was designed…

\[scarcasum\]

Collected metadata like keyboard speeds, page timing, lots of things
Collected 250 responses across both Prolific and MTurk, including a “Please don’t use AI” situation.
Asked an AI agent to also fill out the survey to provide a baseline.

Recommendations:

Do not use MTurk.
- 84% of MTurk responses were LLM
- vs 8% on Prolific
They tracked what participant copied and pasted
- If they copied text from the page, and if they pasted text that was not already present. Also if they pasted text that they did not type.
- This is the main way to differentiate ….
Could we we disabled copy/paste?
- MTurk users re-enabled it or typed in gibberish
Asking users on Prolific to not use AI actually worked
- 0% used AI
- Agents were also less willing to fill out the survey if it said not to use AI on the survey.
Asking for self-report AI usage
- Agent is happy to state that it took the survey
Keep reading the answers you get
- Detecting AI by hand is possible. It is not the content, it is the style
Consider using open-ended attention checks
- Distinctive answers by AI. For example “what is the shape of a ball?” Humans say “round”, AI says “spherical”

These methods work now, but are unlikely to last. AI is changing quickly.

It is important that as a field we work with other researchers to solve these problems. Individuals cannot solve this alone. Data quality checks need to be discussed in paper, because we cannot assume good data for online surveys any longer.

Presenter: Wellington Esposito Barbosa, The George Washington University

Online surveys are prevalent in data collection. But detecting fraud in such studies is very hard. Researchers have to know what real data looks like and what fraud looks like.

Research papers surveyed are coming from non-security communities and is not necessarily being framed as “fraud” the way security would frame it. Researchers often discover fraud while doing other data processing or review. This can be very expensive if the study must be terminated or abandoned.

RQs

What is research fraud?
- No consistent definition among papers.
- Some definitions:
  - Intentional data fabrication
  - Careless and Insufficient-effort
  - Automated, AI-mediated
- Fraud behaviors found
  - Multiple accounts, duplicate responses,
At what stage in research are researchers trying to stop fraud?
- Mitigation strategies are being implemented late stage in cleaning data or were very fragile like asking attention checks.
- For example if a participant fills it out twice: could be blocked based on IP address.

High level point: similar to other aspects of security, we see researchers selecting protections that may not align well with the threats.

Q&A

How can we setup an early warning system for when survey detection methods for bots start being less effective. The publication system is far too slow, and many people need to know these.
- It is a concern that if a “how to avoid survey fraud” was put together, it would likely only be good for a few months.
- We are trying to build a survey system that has security features built into it. And then make it easier to put things that work into it directly.
Double problem: participants are using AI to fill out surveys AND researchers are using LLMs to simulate users in some research. How have we already reflected on the ethics of the means of identification of fraud?
- There were no real discussions of ethics. There were some concerns about how methods were detrimental to reaching specific communities, but it sounds like this was discussed more in terms of research integrity and getting data from a range of users being important.
Should we all go back to in-person surveys? This will slow research. But given the speed of paper submissions, maybe slowing down is a good thing.
- There are pros and cons to that. But we were surprisingly pleased by how effective current mitigation are.

Session: GenAI Privacy & Risks

Presenter: Jonah Bellemans, DistriNet, KU Leuven

Kami was busy and missed most of the talk.

Understanding U.S. Users’ Security and Privacy Transparency Needs for Consumer-Facing Generative AI

Presenter: Jiaxun (Messie) Cao, Duke University

Users have poorly informed mental models about how GenAI is using their data. This includes things like company chat bots.

Most transparency information is expert-facing and aimed at explaining model capabilities and how it uses data.

RQ’s

Do users consider S&P concerns when picking a model?
Current practices and challenges
User preferences

21 interviews with users and 20 designs.

Initial adoption was initially driven by non-S&P factors. Instead popularity, utility, and price were most considered. S&P was a desired adoption feature, but hard to get good data about so not considered.

S&P uncertainty became a larger concern based on the content of the planned discussions with the AI.

When people did read privacy statements, they see it as lacking credibility (8/21) which means they did not trust it. This is a very similar result to what general privacy policy research on how people read them. From my memory: “privacy policies are designed by lawyers to make it hard for me to understand and beneficial to the company.”

User wanted independent evaluations of the privacy policies. Somewhere they could go to compare privacy practices and that had the skills to properly read the practices.

Kami’s thoughts: Consumer Reports in the USA does now include commentary on the privacy practices of some products. Not all, sadly, but their comments on baby monitors regarding privacy are rather amazing.

Presenter: Jennifer Vander Loop

When a loved one dies, those left behind may want to be able to talk to the deceased as part of their digital legacy. Some older people are starting to consider their wills and considering how they want their data used after death, including how they do or do not want to be emulated by a Generative AI.

People were interested in digital resurrection, but less ok with the idea of immortality where such things were used indefinitely.

Ran a survey on 900ish people on Prolific. Looked at what people wanted for themselves and what they wanted for others that had passed. Many people also thought about how creepy this is. Disconnect between teh voice, person, and after-death content. It may sound like them but the content might not match because it did not think like them.

Participants came up with many possible harms. Scams were a large one, they did not want a loved ones voice used to scam others. Also harms to the person’s reputation, and social standing. There is a risk the company might use this for their own befit, such as suddenly getting a loved one to market to you. There is also the risk of non-aligned models where there are two versions that were made with different models. The model itself could be held ransom from the family, especially if an emotional attachment had been made.

Session: Security Advice & Education

Advice given to users can be straight forward, but lead to lots of extra questions around how to opperationalize and understand the advice. There is a difference between being able to find the advice, and the ability to follow it because it can be too complex and make them overwhelmed.

Experts view themselves as a good source of advice, which is partially true, but they are not necessarily good at providing advice for users that is easy to use.

Usability survey used with about 350 participants.

Designing an IT security advice usabilty scale. Started with the , mildy modified to Information Security Policy Usability Scale then resulted in a Adaptation for consumer Advice. The final scale is similar to the SUS in that it is a short 10-question scale.

Older people perceived the advice as more usable than younger people did.

Normal advice like sentence length had no positive advice on usability. But there was some improvement from having clear steps, low tech words, and context alignment.

Consider adjusting advice based on the audience.

Presenter: Farida Chowdhury, BRAC University on behalf of Utsho Das and Argha Pratim Sha

CyQured board game was developed for cybersecurity education.

Security at home is a human decision problem. Users have many devices, no security team at home, and traditional training does not provide practice. So how are home users going about learning about security.

Related work

Many others already on

The game was modeled off of Monopoly. Used STRIDE in a simplified format. There are 16 device cells such as laptops, router, NAS. Placement is also not guesswork.

Conducted a pre-post evaluation with a security knowledge test before and after an hour of group game play. Some participants were also interviewed.

1 hour of play did produce substantial knowledge gains. People with pror security experience did gain more, but everyone learned.

Nice to Know You’re Not Alone: Co-designing Community-centered Online Safety and Privacy Education with Librarians

Presenter: Florian Schaub, University of Michigan on behalf of Tanisha Afnan

People struggle to stay safe online:

Lacking privacy and security literacy
Do not seek help of security experts

Pror work at PETS:

Found that people who are trying to help others, such as librarians, have challenges moving past intro sessions to get people to implement help

In this work, focus was on helping librarians in doing community engagement.

Conducted 4 participatory design workshops with 14 librarians. Participants had 5-20+ years of experience teaching a range of workshops.

Privacy and safety are sometimes taught specifically, but they are often part of a broader digital literacy. They struggled a bit with how to balance conceptual issues with practical skills. This group also has limited resources, such trainings are only one part of their job.

There are many existing educational resources out there that are often diffused and hard to find. Paper has long list of existing educational materials. Existing resources were not easy to just adopt. Many were aimed at experts. Many had loads of jargon. There are also limited resources in non-English. They also wanted to print resources and many are not print-friendly. So most of the time these materials were used to help inform the teacher rather than the learner.

Co-designed solutions for learners:

Interactive games came up
Take-home worksheets and practice exercises
Printed handouts and low-tech materials
Non-class based educational offerings
- Like a monthly community challenge
More need for professional development
- Kami: reminds me a bit about how tackles the issue of teaching Computer Science in schools and taking the view that educators are good at their job, what they need is assistance in making a plan and organizing resources.

Kami note: reminds me a bit of the project that supports older adults in Canada.

Security versus Productivity: A Case Study of Email Management Practices with Job Task Analysis

Presenter: Philip Shumway, University of Tulsa

Cybersecurity policy is written at the higher levels of an organization. That policy then becomes tasks for lower level employees.

Goals:

Re conceptualize security as embedded work
Identify and map task-based tension
Understand how work done

Task Analysis is derived from Job Analysis. Jobs are broken into tasks, and tasks are broken into attributes.

Looked into a case study of a medium sized publisher that had fallen for a phishing attack. So they hired a full time security person to help them be more secure and do security training. Researchers identified a large number of tasks that employees do, broke them down to a smaller set, and then had the employees rank the tasks.

Employees do not see security and productivity as the same. Employees saw getting it wrong more serious for security.

People who self-report having more external emails also spend more time on security.

One interesting thing about this work is the division of employees based on their security context. People have different security needs if they get lots of external emails, than if they did not.

Kami note: Reminds me a bit of the paper which also asked employees to rank security tasks as a way of communicating culture with others.

Symposium on Usable Privacy and Security (SOUPS 2026) - Day 2

Tue, 25 Aug 2026 03:08:19 -0400

Welcome to the happening in Hannover, Germany and being hosted by CISPA.

Blog posts:

Session 1: Privacy Notices, Permissions & Labels

Presenter: Sophia Walsh, University of Bristol

Do the short data notices on app stores properly inform users?

The study looks at users ability to understand the labels in app store notices. They used a card matching approach where they gave users popular apps on one set of cards and a print-out of the short privacy notice icons on another. They then asked the participants to try and match the two sets while doing a think aloud.

Unsupprisingly, they find that users do not know how to map the permissions information to the features that exist in apps. They also find that the information can be overwhelming and causes users to give up.

Presenter: Verena Winterhalter, LMU Munich

More than 372 apps installed on a given phone.

Each of these apps has permissions that control how it can collect and use user data. Users’ awareness of these permissions is poor to start with and can get worse with time since permissions can change over time.

Studied users understandings of what permissions that the user has installed and uses.

Recommends:

Focus on permissions where user has agency
Improve visabiliity & reviwability of installtime permissions
App usage recency as indicator for suggesting app deletion
Different intervention approach per type of misconception

Presenter: Nina Taft, Google

It can be challenging for developers who are building an app to get privacy right. They use libraries, but may not know what those libraries do. But users sometimes have allot to say about the topic in reviews.

This project attempts to give developers personalized reports about their app.

Study is multi-stage. Stage-1: survey of developers about their concerns, privacy perceptions, and odds of taking action. Stage-2: showed them a personalized privacy report. Stage-3: measured their understanding.

The report:

Top 3 privacy concerns from their users’ reviews
Focused on one of those concerns, showed them Volume and Trends
Peer benchmark - compared their app to their peers
Emotions - emotion words

Studied people who work in the privacy field, or at least someone who can make decisions involving the privacy design of the app.

Asked developers about the approaches they have for lookimg at their own reviews, particularly privacy-related. Some read the 1 star reviews. Some have software that reads and summarizes reviews, though not specifically for privacy.

76% of respondants found it useful. The remaining 24% had a “I already knew that” response. 69% say that they are likely to do something about the user concerns after seeing the report.

Session 2: Security Operations & Practitioners

Presenter: Souradip Nath

Looking at how security operations are using LLMs. The reasearch used Reddit to understand usagage. Used mix of Qualtiative and Quantitative.

The AI vender landscape is wide and growing. But awareness of them are fragmented.

Security focused LLMs integrated with Triage and Response while general purpose were more for code writing.

There was a large amount of range in autonomy granted to the agents.

Developers are finding LLMs to be very helpful in doing things like digging through logs and alerts. Agentic AI, in particular, was seen as valuable to find information and quickly answer questions like “does this log line happen every time a user logs in, or is this a new alert?” That said, they were concerned about the accuracy of the tools.

Lots of concerns about how much agency should be given to the tools. The costs of the tools also came up as a serious issue, equating the costs to things like a full time employee salary.

Presenter: Miuyin Yong Wong, University of Maryland

Are LLMs making people more efficient? Are they benefiting productivity of cybersecurity professionals? This field particuarly susceptible to burnout.

RQs (terse version):

What policies and guidance for LLMs are organizations providing
What challenges and risks when using LLMs
How do peer perceptions impact views

Conducted interviews 28 participants from 26 organizations spanning 16 job roles.

The current landscape

Quite a range of policies avaiable. Many did not have a policy or had a relaxed or generic policy. A few had very restricted policy, such as not allowing training on local company data. Participants mostly relied on exsisting data loss measures. Most of the burdon for being safe falls on the individual to not do the wrong thing.

Barriers to adoption and mitigations

Issues like data leakage, hallucinations, lack of confidence in promoting skills, and wasting time.

People were also afraid of other people using the AI. Concerns that less experienced people would use the AI poorly, or loose the ability to learn.

“What will others think of me if I use AI?” The cybersecurity community is an open community that shares knowledge, but AI may be changing that. People are worried about what others will think about their AI use which is causing a lack of communication about it.

Presenter: Faisal Haque Bappy, University of Maryland Baltimore County

Did a coding session using thinking aloud. Divided participants based on the date they were trained, before AI, during the AI boom, or were native to AI use. They also did an interview and a post-task reflection.

the good news is that AI ouptputs are treated as reference rather than a final code state. People talked about the AI like a junior collegue, but they also didn’t explicitly make security requirements to the AI.

Presenter: Katharina Barlage, LMU Munich

DNS converts human-friendly URLs like “vaniea.com” to an IP address (205.196.221.231) which is where the computer establishes a connection. A successful DNS attack can route traffic to somewhere else. Or it could stop users from visiting the site. Having security on DNS servers is vital, but not universally deployed.

DNS “just works” so no one cares untill there is a problem. So upper management does not give resources. Security is often assumed but not assessed. Uptime is a seroius concern and can outrank security.

Some DNS opperators are buying software to run, and assuming that security is there because it should be. They do not assess. Because everything looks fine there is no percieved need to go adjust things. The danger is also abstract and not seem real because there have not been any DNS incidents that they know of. But if they adjust DNS wrong, everything will break, so the short term risk seems more risky. Turning on security is similar, turning on security seems risky and not worth that risk.

“I didn’t know I would be this excited not to be scammed.” Exploring Emotional and Behavioral Responses During Phishing Attacks

Presenter: Raphael Weidhaas, Aalto University

People are still clicking on phising despite lots of training. There is limited research on how phishing impacts the emotion state of the person who gets the phishing.

Lab study where users were put in an office setting. They were given a set of tasks. After about 45 minutes they were sent a simulated phishing email. It was mildly associated with a task, but had some clear indicators like three ‘o’ on zooom. Followed by an interview.

22 participants identified the phishing, 7 clicked, 8 avoided but didn’t realize it was a phishing. 4 fell for the phishing and entered data. The clickers tended to think they had made an error and were trying to correct it, but did realize that it was a problem. Phished people felt that they had solved the problem and felt positive afterwards.

People who fell for phishing had the most positive emotions. Detectors were the most annoyed.

Presenter: Martin Dechant, University College London

How does having Social Anxiety impact how people process and think about phishing.

People with social anxiety are worried about situations were they are around large groups of others where they might stand out or be criticised by those peers. 4.7% of children, and 8.3% of adolescents have social anxiety globally.

Asked a group of people about a recent social engineering attack that they had experienced. Scammers exploit trust and desires by creating a sense of familiarity, then manipulating with the promises of materialistic or financial gains. Perpetrators weaponized any shared material. Perpetrators fabricate a crisese or urgent situation.

Studied three scams where participants experienced a scenario as a role play.

Job Scam (Trust and Desires)
Extortion (Compromising material)
Romance Scam (Empathy and Urgency)

Participants did not want to reach out for help due to structural barriers. Social Anxiety people didn’t want to talk about it, they were worried about judgement, assumed that others would laugh.

Scammers used language to give affirmation that the social anxious person craved like “you are great” and “you are amazing”.

Quantifying Risk Perception and Scam Response Among International and Domestic US University Students

Presenter: Elijah Bouma-Sims, Carnegie Mellon University;

International students may be more vulnerable to scams for reasons like: unfamiliarity with US legal, lingustic, and cultural norms, pressures for post-graduation employment.

Surveyed about 1000 international and domestic students from 5 universities. Each person saw a scenario describing a scam from a set of sources like phone call, sms, email. They were also asked about prior scam experience.

Experiences with Digital Scams Post-Incarceration in the U.S.

Presenter: Yael Eiger, University of Washington

Incarcerated people are more commmen than you might realize. It is in the millions. 7.6 millions times a year people are sent to jail because they cannot make bail and must stay there till their case is resolved. More people are Incarcerated than are convited of any crime.

Research looks at how previously Incarcerated people experience scams. The research is interesting in pointing out what it is like to experience all of technology all at once, including setting up accounts. This population is also at great risk from things like parking tickets, which makes scams more scary.

This group also have to register their existance, so they are in public databases which scammers then use.

Session: At risk and vulnerable users

Usability Determines Safety for At-Risk Users: Evaluating Hidden Device Detectors for Intimate Partner Surveillance

Presenter: Akhil Polamarasetty, University College London (UCL)

Trackers are now cheap to obtain, and easy to buy. There are also lots of detectors. Physical dtectors use things like RF Signal Detection, Magnetometer, Lens Detection. There are also mobile apps that do things like scanning wifi and bluetooth.

The researchers did several rounds of testing of various devices, such as testing in an anechoic chapber and dark room to see what the detectors could do in opitmal conditions. Then had users try them in a fake living room.

Users searched the room in two ways:

Intuitive: looked for sply-looking objects like clocks
Systematic: Divide a room into a grid and looked at most objects. This tactic was more common for people with law enforcement background

Having a physical detector tended to make people feel more vulnerable, even though they were in a safe lab space.

Presenter: Tarini Saka

Recovery from serious security issues is challenges. Many people turn to online support. But the online communities themselves have minimal security.

This work looks at the context of labor abuse. For this group Human Resource departments, which are the reporting destination, may not help and re-traumitize instead. Instead people try to find groups to communicate with safely about what they can do.

Labor abuse survivers form networks across platforms where they are forming an audience. Compare them to activists who have nation-state abilities. Labor abuse victim face locally powerful adversaries that have financial and direct ability to impact the victim, but they have less ability to impact the technology world.

This is an interview and thematic analysis study.

Many people reached out to understand if their situation was indeedproblematic or if they were just “crazy”. The researchres group into four types of communication groups: feed-based groups, publicly visible threads, Messaging groups, ???

Presenter: Sohana Akter, University of Texas at San Antonio

Instant messsaging is used for a large parts of personal and professional life. The same is true for the Blind and Partially Sighted users.

Blind and low vision people may have further issues like tryingn to understand an image, others might hear a screen reader, and then there are privacy controls….

Just moving a mouse into some spaces can activate things, but a blind user has no way to know that will or has happened. Tags like “last seen” are intended to be seen at a glance, but that is more challenging for partially sighted.

Some information entry is irreversible, which is more challenging when using a screen reader which may not properly explain what information is entered or why.

[]“Don’t Let Them Get To You”: Understanding the Role of TikTok as a Source of Support for Cyberbullying Victims]( )

Presenter: Daniel Zappala, Brigham Young University, on behalf of Saba Iqbal

Cyberbulling is increasing as a problem in the US.

Prior work shows that sorting through security and privacy advice is challenging and users strugggle to prioritize it.

This work looks at the cyberbulling advice that exists on TicTok and how accurate that advice is.

Looked through lots of TicTok videos and found, that there were roughly 5 categories of advice:

Coping strategies
- Things like: don’t respond, block and report, take a break, comfort the bully
Emotional Support
- Reassuring that bulling is not acceptable, not true, projection of bully insecurities, empower yourself.
Mental Health Advice
- Set boundaries, don’t take it personal, exercise, journaling, get sleep,
Spiritual Support
- Support based on faith things like: Turn the other cheek, god is watching over you,
Seeking support
- Guidance on where to turn to. Like get evidence, contact law enforcement

Survey was then run where asked about experiences, and asked them the rate the content of the videos (text transcription). Rated baed on: comprhensibility, efficacy, actionability.

Most advice was rated as comprehensible. Most advice was considered effective. The spiritual category was rated lower for being helpful. About 1/3 of advice was considered actionable. Confrunting the bully, for example, were not seen as actionable.

Reproductive Security & Privacy Advice on TikTok after the Overturn of Roe

Presenters: Harshini Sri Ramulu and Rachel Gonzalez Rodriguez, Paderborn University

The overturning of Roe vs Wade in the US lead to many women suddenly loosing reproductive care and requests for information about how to protect themselves.

Study looked at TikToc videos that gave advice on protect yourself as a woman in the US. The covered issues like how to not track mentration (delete period tracking apps). There were lots of contradictory advice, such as using apps in Europe or not using apps because they were in Europe. Some advice also suggested creating a fake data.

Symposium on Usable Privacy and Security (SOUPS 2026) - Day 1

Mon, 24 Aug 2026 07:02:13 -0400

Welcome to the happening in Hannover, Germany and being hosted by CISPA. The SOUPS community is made up of researchers that care about understanding and improving peoples’ interactions with security and privacy technologies. Research in this area represents a very wide range of cybersecurity and privacy topics with a focus on people.

Blog posts:

Authentication & Credentials

Presenter: Alexander Krause

Updating passwords is something many people do. But do websites always support password resets and how easy is it to reset passwords?

Users have many situations where they might want to change their password:

Maybe they just want to feel more secure
Maybe they lost their password
Maybe they want to make it more challenges

Goal of the work is to look at three questions:

How do websites implement password update processes?
How doe sit go wrong?
How can we do it better?

The researchers looked through the password reset processes on a large number of websites. There is also a that they tested.

The path is long - finding and resetting passwords is not easy. 3-6 clicks. And the setting is hiding in many differently-named locations.

Only 1 of 111 check the new passwords against leaked ones
Only 2 have a generator built into the site
Only 9 can fil the form correctly with a password manager

There is not much feedback either. Some feedback only shows for 1-2 seconds. Some give no feedback at all. Leaving a user to ask “did it work”? 34 of 96 change flows and 46 of 109 reset flows sends no email.

Note that the lack of email is bad for abuse situations as it means that an attacker could change someone’s password without their immediate knowledge

Recommendations

End other user sessions when a password challenges
Always send an email which includes a “if you didn’t do this, how to fix” link
Make the W3C password URL work by default

Passkeys in the Wild:

Presenter: Michael Clark

Passkeys are a phishing-resistant authentication built off the FIDO2 (WebAuthn + CTAP) technology. It is built on top of public/private key technology. The general idea is that the client computer creates and protects a private key. It then gives the public key to websites during authentication setup. The client prooves its identity by signing a challenge using its private key. The workflow often also involves a user providing a biometiric to the local client device.

Passkey usability dpends on UX consistency - users spend most of their time on other webisstes, so they expect your site to work like all the other sites they already know. (Jakob’s law)

Evaluated 111 websites, looked at user paths through the websites. Looked at 28 different features and used it to measure similarity.

Discovery
- Promotion, priority, educatinal resources, naming of the feature
Setup
Usage
Deletion
- Super important for recovery after attack by in-home attackers or malware

Found that there was a lack of full support for the full lifecycle.

Only 3% use PassKeys during password recovery
Only 23% told the user to also delete a deleted passkey from their local passkey manager

Missmatch in that websites let you use passkeys, but non gave information on how to create one.

Mobile apps are not letting users use a passkey that was originally setup on the website.

Presenter: John Sadik

Password managers are super useful, but it can be challenging for users to use them.

Kami got distracted….

Presenter: Sushanth Ambati, Rowan University

Digital identity management can be challenging. Self-soverenty where users are fully in charge of their own identity management can be empowering and allow great things like flexibility and freedome from government decisions. But it also makes it harder for the user who now has to handle issues like backup, loss, storage, and all these issues.

Contributions:

Baseline Evaluation of 128-bit Keys
Partial Key Recovery
Cognitive-Driven Key Recovery

It would be wonderful if people could memorize 128-bit keys which are massively better than passwords. But well, human memory does not work like that. Humans are very good at memorizing other information though, like location of objects in spaces, sets of images, and lots of others.

Idea is to create a type of graphical approach to key recovery. An imporant realization is that memory is not normally completely wrong, it is normlly partially wrong and partially right.

Session 2: Families & Shared Devices

Presenter: Ann-Kristin Lieberknecht, Goethe University Frankfurt

Manging privacy and security is a challenge for parents and there is quite a bit of fragmented reserach on this topic.

How is knowlege about parental privacy actually translated into intervention design.

Reviewed 22 research papers that looked at active interventions by parents on the topic of security and privacy.

Found a very wide range of interventions ranging from parents to schools, to how the delivery happened or what content was in it. The research in this area is very fragmented and similarly the design of intervetions is quite fragmented.

Design space tensions

Supporting action without oversimplifying
Individual responsibility with distributed family expertise

Families with children grow over time whith knowledge and expertise changing across time.

Scalablility Constraints

Depth, adaptation and scalability hard to do all at the same time

The research field is better developed in problem articulation than in intervention validation.

Presenter: Franziska Roesner, University of Washington on behalf of Meghna Gupta

The practice of sharing photos, videos, or other information about their own parent’s own children via online platforms. They share to:

Document memorieis and milestones
Showcasing their parental identity
Seeking community support

There are also online harms and risks to sharing:

Offline harms: bullying, embarasment
Online harms: identity theft

Non-parents also create and share information about children like Grandparents, extende family, and professional caregivers (coaches, teachers, daycare workers)

Research focus is on in-home childcare workers. This group forms a deep bond with a family and is very involved.

Semi-structured interviews with 8 parents, 7 in-home childcare workers. Goal was to understand prevaleng norms, perceptions and negotiations that occur.

In-home childcare workers often take photos:

Children request that photos be taken
Parents ask for photos
- Documentation request
- Documentation as proof of work
Childcare workers want to record moments

Privacy norms are not normally explicitly stated. Sometimes contracts will state about sharing, but parents are downloading a template of employment contract and maybe not thinking about it. More experience childcare workers were more explicit about establishing boundaries. Most of the workers had to interpret what the parents wanted.

No consistent norm for what to do when a child care worker stopped employment. Deleting photos requests can be taken personally because the childcare worker is deeply involved in raising the child. Who owns these photos is also confusing and who gets to judge.

Presenter: S. Shanza, University of Waterloo

Immigrants are intersting to study. 1 in 4 Canadians is an immigrant, many from “collectivist” cultures where objects, including devices, are viewed a communally owned.

In collectivist households sharing of devices is normal as many things are owned by “the family” as opposed to by an individual. But many devices are designed to be owned by a single user. Or at least to have one user behind one each account.

Interviewed 18 families (at least one parent and one child) who had immigrated in the last 20 years. 41 participants in total.

Device sharing is used as a form of control over children behavior. Because the device is shared the parents can see. Devices are meant to be kept in a “clean” state. Even parent devices are kept like that.

Device sharing does cause stress. Children described being stressed about having fights about content with parents. And parents were stressed about things like having their own work disrupted because tools or internet access were removed so students would not have them. Also about how to keep children safe and give them good vlaues.

Shared devices also mean shared security practices. If anyone, even children, clicks on the wrong thing and gets malware, it impacts the whole family because all the access is happening via a shared devices, even banking.

While tools like profiles exist to support different users on one device, they were hidden behind lots of settings so they were not used.

Presneter: Darya Zarkalam, Paderborn University

Offline second-hand shopping has existed for a long time. These platforms have moved online, for example eBay or Vinted. These markets are unique in that they provid features like messaging, ratings, payment and shippment. But ultipmately users decide how to complete the transactions.

Scammers are definitely use these platforms too.

Study looked at how people think about and use these platforms from a security and privacy perspective.

Users security and privacy concerns
Users perceptions
Users strategies for avoiding loss
How users build and establish trust

Participants were concerned about security and privacy. They worried about sharing data about themselves like their phone number.

Users tried to protect themselves. They tried to use payment methods they considered safe like cash, or paypal where there is documentation. Integrated payment features were seen as reducing odds of financial loss, platform can be involved if something goes wrong.

They avoid sharing sensitive details. They shared data progressively as it was needed.

Reputation signals were used. Such as not talking to people who had lower reputation scores.

Platforms provide little guidance

Keynote: Usable Security in Practice - How Tuta Makes Email Encryption Usable

Presenter: Carmela Troncoso

“Turn on privacy”

provides end-to-end encrypted email services using quantum-secure encryption. It enables automatic encryption between Tula users. For communications with non-Tula accounts a password is shared in advance and then can be used for all future communications.

Tuta aims to enable Digital Sovereignty or just alow people to have control over their own digital life.

Seamless encryption is more than just who possesses the private key. Issues like groups communicating (group keys).

Keeping up with tech is also non-trivial. If only a user has a key (or the password to unlock the key) then by definition the company does not have it. That makes it hard to do things like update encryption algorithums to the latest quantum-protected ones. Now it is only possible to update the algorithums when each user logs in when the keys become available.

Legal protections are also a serious concern. Tuta is hosted in Germany allowing them to benifit from German laws and the GDPR which gives users legal protection not just technical protection.

Common Referencing Errors

Fri, 05 Jun 2026 15:51:43 -0400

As Technical Paper Co-Chair for , I lead a group of volunteers that checked references in accepted but not-yet-published papers. I reached out to authors about the serious issues we observed and how they came about. I have also been speaking to members of the community about the process we took this year and listening to the experiences of other conference organizers. Below I detail recommendations based on my observations. Each recommendation is paired with a “why” where I talk about what we observed.

Outline

Reference storage

Store references directly in BibTeX

Store references such that all the data is correctly stored in a stuctured format, ideally BibTeX. At minimum store references in a format that can be directly exported to BibTeX (e.g. Mendeley or Zotero) without having to go through an intermediary that has to re-interpret the reference content: like an LLM or a junior researcher.

Keep in mind that conversion to a BibTeX file requries some modification of references such as identifying authors’ first and last names seperately, and ensuring that some capitalization, such as DNS, is properly preserved.

Why: Several possible hallucinations at SOUPS this year were caused by an author storing their references in some other format such as Word or Excel and then day of submission trying to move all those references over into BibTeX. Due to time limitations they used approaches like asking an LLM to do it or asking a very junior researcher to do it for them. We observed that such reference copying resulted in fabricated-looking references.

Keep separate BibTeX files based on reference provenence

Know which references have been carefully checked and which have not. One option LaTeX provides is having multiple BibTeX files. Consider using this feature to store the references that have been carefully checked seperate from those where checking has not yet happened. Also consider using it to seperate out references that have been blinded for submission.

Why: It is good practice to record the papers read somewhere, and it can be time consuming to clearly check all aspects of a reference that may never be used in a paper. As a result, people use tools like Zotero and Mendeley which can automatically collect bibliography information and store it quickly.

These tools are amazing, save time, and ensure researchers have a fast-to-create record of papers they have seen. But these tools very often fail when it comes to creating a full and correct reference. They can omit authors, leave out venues, or record the wrong location for a conference. They are also less effective at capturing things like which chapter of a book the researcher is referring to, or the exact page number being referenced.

We observed many of these errors at SOUPS 2026. Some communication with authors suggest that tools like Zotero and Mendeley combined with not double checking their output is to blame. So please use these tools with care, and to save time consider using seperate BibTeX files for auto-collected references and hand-cleaned references.

Diff any changes AI makes to paper text

Tracking of changes made by authors is already a supported feature in tools like Overleaf and Git. Because LaTeX is code-like, many labs are already storing their raw paper LaTeX in repositories that support tracking and visual representations of changes. Even Word has some limited ability to track changes.

We recommend that if an AI is used to do “minor” things like “just clean up the text a bit”, that authors consider doing a local checkin before/after the change and review any tool-provided diff of the versions. It is vital to make sure the AI only changed the parts of the paper it was instructed to. The easiest ways to do such checks is to automatically compare the old and new text with a tool that creates a visual highlight of the changes. Doing so makes it easy to skim to see what was changed.

Why: Many PhD students at conferences talk to me about near miss situations where they asked GenAI to make a small change in one part of the paper only to find later that it had touched other parts like swapping out part of their bibliography references. These cases left the students supprised and a bit shaken that the AI would do that. The changes they described were also well outside the scope of what they had instructed the AI to do. A simple skim of a visual diff, or a list of line numbers with changes, would have easily caught these cases.

Tools: I would love it if Overleaf allowed us to register GenAI as a separtate user and mark all changes it makes as needing approval. It would make it much easier for a student to review the changes, learn from the changes, and spot changes that they did not ask for with the option to easily reverse them.

Openly discuss GenAI usage in labs

Generative AI is being actively used by all types of researchers, it is also being actively added into tools used by researchers, sometimes with very little labeling. We recommend having open discussions within lab groups about what kind of GenAI tools people are finding useful, how they are using them, and how they are ensuring accuracy of any output.

There is currently a view that Generative AI can be used responsibly. Regardless of if you subscribe to this view or not, odds are some of the other members of your lab or larger research groups do hold this view. I believe that it is important to have open dialog about what responsible AI usage looks like and actively discussing how output is being reviewed. Outright banning GenAI usage may be causing lab members to use it in secret or not be willing to openly discuss tools that maybe use GenAI. It is important as a community that we ensure open and supportive dialog within labs on this topic.

Why: In all cases we handled this year, the potential fabrications were a supprise to supervisors. Some firmly believed, before our communication, that GenAI was not used in their paper construction at all. When talking with students at conferences there is often a tone of “well I only used it for X” with an implication that this was a small immaterial usage that need not be disclosed. We expect that this attitude combined with knowledge that the supervisor does not support GenAI usage, may be leading to secret usages of GenAI.

Write clean, accurate BibTeX

Below are high-level suggestions to keep in mind when cleaning references.

Paper PDF on the publisher site is the official correct information

Officially the correct reference is always the information on the paper PDF as downloaded from the publisher’s website. If there is a conflict between the paper PDF and the publisher metadata, the PDF is considered to be the more accurate. When providing a URL to a paper, preferably use the DOI that links to the publisher version, or a link to the publisher’s site. Avoid linking to less-canonical sites like the author’s website, or Research Gate unless there is no other option.

Why: The metadata on places like DBLP, and Google Scholar have a high quality mix. We saw several cases of authors taking their bibliography information from these sites, but that information did not match the actual paper. We also observed cases where the correct DOI was linked to, but author names had been copied from Google Scholar and did not match the PDF the link lead to.

Many publishers offer a BibTeX citation download directly, this is often, but not aways, accurate. It is advisable to always start with the publisher-provided BibTeX and then check that the information matches the paper PDF and nothing is missing.

arXiv: Include version numbers when referencing arXiv

arXiv gives several DOI number options for citing a paper, including to the paper in general AND the individual versions. Make sure to use the DOI that includes the version number.

arXiv:2408.08967v1 <- use
arXiv:2408.08967 <- avoid

Why: An arXiv paper is not peer-reviewed, so it can change quite significantly between versions, including swapping out the title entirely or changing the author list. We saw some papers get flagged by the citation checker because the reference information was from v1 but the link didn’t specify the version, so the citation checker looked att the most recent version (v2) and concluded that the reference did not match and might be hallucinated.

Train students in BibTeX author list and title formatting

Author lists and titles are the two parts of BibTeX that require the most manual reformatting when copying content off the Web. Students need to be trained in how to correctly do the necessary conversions.

Authors in BibTeX format need to be formatted: surname, first name and surname, first name. Critical here is the ordering of surname first, and the use of ‘and’ between author names. Other formats produce hallucinated-looking output.

Titles of papers need to be adjusted to ensure preservation of upper and lower case. BibTeX will default to sentence case, but that will cause words like “URL” to be converted to “url”. The easiest solution is to use double curly brackets around the correctly formatted title (see example below).

Why: Graduate students seem to be turning to GenAI to help them convert references into BibTeX partially because they do not know how to write BibTeX themselves and they do not know what correctly formatted BibTeX looks like.

We saw cases with missing or mangled authors where the cause turned out to be poor BibTeX author formatting of a real and otherwise accurate BibTeX entry. See the correct and wrong examples below. Note the impact of BibTeX formatting on the resulting capitalization and author names.

Correct

@inproceedings{althobaiti2021chi,
 author = {Althobaiti, Kholoud and Meng, Nicole and Vaniea, Kami},
 title = {{I Don't Need an Expert! Making URL Phishing Features Human Comprehensible}},
 booktitle = {CHI '21},
 year = {2021}
}

Rendered in IEEEtran format:

K. Althobaiti, N. Meng, and K. Vaniea, “I Don’t Need an Expert! Making URL Phishing Features Human Comprehensible,” in CHI ’21, 2021.

Wrong

@inproceedings{althobaiti2021chi,
 author = {K Althobaiti, N Meng, K Vaniea},
 title = {I Don't Need an Expert! Making URL Phishing Features Human Comprehensible},
 booktitle = {CHI '21},
 year = {2021}
}

Rendered in IEEEtran format:

K. V. K Althobaiti, N Meng, “I don’t need an expert! making url phishing features human comprehensible,” in CHI ’21, 2021.

The above rendered output has several issues:

Incorrect author parsing. In BibTeX commas do not separate authors, the only way to separate authors is the word “and”. Any other way of breaking up authors causes them to be parsed as one author.

Detailed explaination: Because of the lack of ‘and’ in the author list, BibTeX only sees one author. Imagine the name “Simon Jones, Jr” where “Jr” is the suffix, in this case it is short for “Junior” which is used when a parent and child share a name. In LaTeX this should be written like: “Surname, Suffix, FirstName” or “Jones, Jr., Simon”. In the example above, LaTeX has parsed “K Althobaiti” as the surname, “N Meng” as the suffix, and “K Vaniea” as the first name. Thereby producing the singular author: “K. V. K Althobaiti, N Meng” formatted as “abbreviated firstname surname, suffix”

Capitalization wrong. The title has “url” instead of “URL”. Similarly there is a `!’ incorrectly followed by a lowercase. The use of double curly braces in the correct version above prevents this capitalization adjustment from happening.

Full author names. Full authors names should always be recorded in BibTeX. The IEEEtrans format above auto converts first name’s into initials, but other formats like ACM Reference Format shows the first names. It is always better to include first names in full in the BibTeX so that different formats can be applied.

SOUPS | Kami Vaniea

Symposium on Usable Security and Privacy - Day 3

Session: Usable Security Mechanisms & Emerging Threats

Lightning talks

Session: Scams, Fraud & Misinformation

[Source-Level Disengagement: A Usable Security Defense Against Misinformation](Source-Level Disengagement: A Usable Security Defense Against Misinformation)

Q&A

Session: GenAI Privacy & Risks

Understanding U.S. Users’ Security and Privacy Transparency Needs for Consumer-Facing Generative AI

Session: Security Advice & Education

Nice to Know You’re Not Alone: Co-designing Community-centered Online Safety and Privacy Education with Librarians

Security versus Productivity: A Case Study of Email Management Practices with Job Task Analysis

Symposium on Usable Privacy and Security (SOUPS 2026) - Day 2

Session 1: Privacy Notices, Permissions & Labels

Session 2: Security Operations & Practitioners

The current landscape

Barriers to adoption and mitigations

Session: Phishing & Social Engineering

“I didn’t know I would be this excited not to be scammed.” Exploring Emotional and Behavioral Responses During Phishing Attacks

Quantifying Risk Perception and Scam Response Among International and Domestic US University Students

Experiences with Digital Scams Post-Incarceration in the U.S.

Session: At risk and vulnerable users

Usability Determines Safety for At-Risk Users: Evaluating Hidden Device Detectors for Intimate Partner Surveillance

[]“I feel as though my privacy is being violated”: Privacy Risks and Barriers in Instant Messaging for Blind and Low-Vision Users]( )

[]“Don’t Let Them Get To You”: Understanding the Role of TikTok as a Source of Support for Cyberbullying Victims]( )

Reproductive Security & Privacy Advice on TikTok after the Overturn of Roe

Symposium on Usable Privacy and Security (SOUPS 2026) - Day 1

Authentication & Credentials

Passkeys in the Wild:

Session 2: Families & Shared Devices

Keynote: Usable Security in Practice - How Tuta Makes Email Encryption Usable

Common Referencing Errors

Reference storage

Store references directly in BibTeX

Keep separate BibTeX files based on reference provenence

Diff any changes AI makes to paper text

Openly discuss GenAI usage in labs

Write clean, accurate BibTeX

Paper PDF on the publisher site is the official correct information

arXiv: Include version numbers when referencing arXiv

Train students in BibTeX author list and title formatting

Correct

Wrong