Shirin Nilizadeh
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Shirin Nilizadeh has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Poojitha Thota
The University of Texas at Arlington
4 shared publications - Sayak Saha Roy
The University of Texas at Arlington Arlington Texas USA
4 shared publications - Abolfazl Asudeh
University of Illinois Chicago, Chicago, Illinois, USA
3 shared publications - Gabriela Mustata Wilson
College of Nursing and Health Innovation, University of Texas at Arlington
2 shared publications - Gautam Das
The University of Texas at Arlington
2 shared publications Krishna Vamsi Naragam
The University of Texas at Arlington
2 shared publications
Latest Publications
- Open Access
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1291-1306
Unequal Privacy: Auditing Demographic Bias Vulnerabilities in Visual Protection Systems
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1291-1306
Open AccessUnequal Privacy: Auditing Demographic Bias Vulnerabilities in Visual Protection Systems
- Open Access
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1667-1683
The Insider's Advantage: Exploiting Automated Privacy Policy Analyzer Tools Through Subtle Text Manipulations
Privacy policies are essential for communicating data practices to users, despite their dense, complex language, which often obstructs comprehension. Automated Policy Analyzer Tools (APATs) and recent advancements in Large Language Models (LLMs) have made strides in simplifying these documents through features such as question-answering and summarization. However, the robustness of these tools against adversarial manipulations remains less explored. This study examines how subtle, organization-side modifications to policy text can mislead APATs. We present APATRA, a multi-level framework for assessing the robustness of APATs against a range of policy-specific attacks designed to manipulate them into generating misleading outputs. We uncover vulnerabilities that can lead to substantial errors in outputs by applying character-, word-, phrase-, and sentence-level perturbations to policy contexts while preserving the original information. Our evaluation spans policy-specialized models (PolicyQA, PrivBERT), state-of-the-art LLMs (LLama-4-Maverick, GPT-3.5, GPT-4, GPT-5-mini, Claude-3.5, Claude-4.6-Sonnet), and third-party policy analyzer tools (AesirX). To validate the subtlety of our attacks, we conducted an IRB-approved user study with 100 general participants and 10 legal experts specializing in privacy law, who assessed adversarial excerpts for grammaticality, contradictions, and logical flow. Their evaluations demonstrate that our modifications maintain readability and raise no concerns. Overall, our results reveal critical security gaps in APATs and underscore the urgent need to enhance their robustness.
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1667-1683
Open AccessThe Insider's Advantage: Exploiting Automated Privacy Policy Analyzer Tools Through Subtle Text Manipulations
Privacy policies are essential for communicating data practices to users, despite their dense, complex language, which often obstructs comprehension. Automated Policy Analyzer Tools (APATs) and recent advancements in Large Language Models (LLMs) have made strides in simplifying these documents through features such as question-answering and summarization. However, the robustness of these tools against adversarial manipulations remains less explored. This study examines how subtle, organization-side modifications to policy text can mislead APATs. We present APATRA, a multi-level framework for assessing the robustness of APATs against a range of policy-specific attacks designed to manipulate them into generating misleading outputs. We uncover vulnerabilities that can lead to substantial errors in outputs by applying character-, word-, phrase-, and sentence-level perturbations to policy contexts while preserving the original information. Our evaluation spans policy-specialized models (PolicyQA, PrivBERT), state-of-the-art LLMs (LLama-4-Maverick, GPT-3.5, GPT-4, GPT-5-mini, Claude-3.5, Claude-4.6-Sonnet), and third-party policy analyzer tools (AesirX). To validate the subtlety of our attacks, we conducted an IRB-approved user study with 100 general participants and 10 legal experts specializing in privacy law, who assessed adversarial excerpts for grammaticality, contradictions, and logical flow. Their evaluations demonstrate that our modifications maintain readability and raise no concerns. Overall, our results reveal critical security gaps in APATs and underscore the urgent need to enhance their robustness.
- Open Access
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 788-804
Taming Data Challenges in ML-based Security Tasks Using Generative AI
- 4 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 788-804
Open AccessTaming Data Challenges in ML-based Security Tasks Using Generative AI
- Open Access
- 27 April 2026
- Published by JMIR Publications Inc. in Online Journal of Public Health Informatics
- Vol. 18, e72465
Assessment of the Cultural Nuances in COVID-19 Vaccine Uptake Through a Comparative Analysis of English and Spanish Facebook Posts in Tarrant County, Texas: Longitudinal Study
Background: Prior studies have identified key factors contributing to COVID-19 vaccine hesitancy, including concerns over vaccine safety, potential side effects, and mistrust in the health care system. According to the World Health Organization, vaccine hesitancy is among the top 10 threats to global public health. Previous research has suggested that vaccine hesitancy is a significant barrier within the Hispanic population, particularly in Texas. Objective: This longitudinal study examined the relationships of daily stance, misinformation, and topics in vaccine-related English and Spanish Facebook posts with daily vaccination rates in Tarrant County, Texas, during 2021 and 2022. The goal was to identify the predictors associated with vaccination uptake and inform targeted social media interventions, with particular attention to the Hispanic population. Methods: COVID-19 vaccine-related English and Spanish Facebook posts from Tarrant County were collected for 2021 and 2022. The study analyzed 12,395 English posts and 1123 Spanish posts. Posts were annotated using GPT-4 for stance, misinformation, and relevant topics, including vaccine availability, safety, and side effects. Category prevalence was compared across English and Spanish posts and across years. Linear regression models were used to examine associations between post characteristics and daily vaccination rates in the total and Hispanic populations. Results: Regression analysis identified distinct predictors of Hispanic vaccination uptake, including encouraging posts ( P =.02) and religion-related posts ( P =.007), which were not significant predictors for vaccination uptake in the general population. A substantial proportion of Spanish discouraging posts focused on vaccine side effects (13/70, 19%) and health system distrust (24/70, 34%), suggesting concerns that may be especially relevant within the Hispanic community. Predictors associated with higher uptake in both the Hispanic and total populations included posts related to vaccine availability ( P =.01), vaccine safety ( P =.006), and misinformation debunking ( P <.001). Conclusions: Posts related to vaccine availability, vaccine safety, and debunking misinformation were associated with higher vaccination uptake. Encouraging posts and religion-related posts were associated with higher vaccination uptake in the Hispanic population, suggesting meaningful cultural nuances. These findings support the value of culturally tailored social media messaging in public health campaigns.
- 27 April 2026
- Published by JMIR Publications Inc. in Online Journal of Public Health Informatics
- Vol. 18, e72465
Open AccessAssessment of the Cultural Nuances in COVID-19 Vaccine Uptake Through a Comparative Analysis of English and Spanish Facebook Posts in Tarrant County, Texas: Longitudinal Study
Background: Prior studies have identified key factors contributing to COVID-19 vaccine hesitancy, including concerns over vaccine safety, potential side effects, and mistrust in the health care system. According to the World Health Organization, vaccine hesitancy is among the top 10 threats to global public health. Previous research has suggested that vaccine hesitancy is a significant barrier within the Hispanic population, particularly in Texas. Objective: This longitudinal study examined the relationships of daily stance, misinformation, and topics in vaccine-related English and Spanish Facebook posts with daily vaccination rates in Tarrant County, Texas, during 2021 and 2022. The goal was to identify the predictors associated with vaccination uptake and inform targeted social media interventions, with particular attention to the Hispanic population. Methods: COVID-19 vaccine-related English and Spanish Facebook posts from Tarrant County were collected for 2021 and 2022. The study analyzed 12,395 English posts and 1123 Spanish posts. Posts were annotated using GPT-4 for stance, misinformation, and relevant topics, including vaccine availability, safety, and side effects. Category prevalence was compared across English and Spanish posts and across years. Linear regression models were used to examine associations between post characteristics and daily vaccination rates in the total and Hispanic populations. Results: Regression analysis identified distinct predictors of Hispanic vaccination uptake, including encouraging posts ( P =.02) and religion-related posts ( P =.007), which were not significant predictors for vaccination uptake in the general population. A substantial proportion of Spanish discouraging posts focused on vaccine side effects (13/70, 19%) and health system distrust (24/70, 34%), suggesting concerns that may be especially relevant within the Hispanic community. Predictors associated with higher uptake in both the Hispanic and total populations included posts related to vaccine availability ( P =.01), vaccine safety ( P =.006), and misinformation debunking ( P <.001). Conclusions: Posts related to vaccine availability, vaccine safety, and debunking misinformation were associated with higher vaccination uptake. Encouraging posts and religion-related posts were associated with higher vaccination uptake in the Hispanic population, suggesting meaningful cultural nuances. These findings support the value of culturally tailored social media messaging in public health campaigns.
- Open Access
- 7 June 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the International AAAI Conference on Web and Social Media
- Vol. 19, 1798-1816
Auditing Yelp’s Business Ranking and Review Recommendation Through the Lens of Fairness
Auditing is critical to ensuring the fairness and reliability of decision-making systems. However, auditing a black-box system for bias can be challenging due to the lack of transparency in the model’s internal workings. In many web applications, such as Yelp, it is challenging, if not impossible, to manipulate their inputs systematically to identify bias in the output. Yelp connects users and businesses, where users identify new businesses and simultaneously express their experiences through reviews. Yelp recommendation software moderates user-provided content by categorizing it into recommended and not-recommended sections. The recommended reviews, among other attributes, are used by Yelp’s ranking algorithm to rank businesses in a neighborhood. Due to Yelp’s substantial popularity and its high impact on local businesses’ success, understanding the bias of its algorithms is crucial. This data-driven study, for the first time, investigates the bias of Yelp’s business ranking and review recommendation system. We examine three hypotheses to assess if Yelp’s recommendation software shows bias against reviews of less established users with fewer friends and reviews and if Yelp’s business ranking algorithm shows bias against restaurants located in specific neighborhoods, particularly in hotspot regions, with specific demographic compositions. Our findings show that reviews of less-established users are disproportionately categorized as not-recommended. We also find a positive association between restaurants’ location in hotspot regions and their average exposure. Furthermore, we observed some cases of severe disparity bias in cities where the hotspots are in neighborhoods with less demographic diversity or higher affluence and education levels.
- 7 June 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the International AAAI Conference on Web and Social Media
- Vol. 19, 1798-1816
Open AccessAuditing Yelp’s Business Ranking and Review Recommendation Through the Lens of Fairness
Auditing is critical to ensuring the fairness and reliability of decision-making systems. However, auditing a black-box system for bias can be challenging due to the lack of transparency in the model’s internal workings. In many web applications, such as Yelp, it is challenging, if not impossible, to manipulate their inputs systematically to identify bias in the output. Yelp connects users and businesses, where users identify new businesses and simultaneously express their experiences through reviews. Yelp recommendation software moderates user-provided content by categorizing it into recommended and not-recommended sections. The recommended reviews, among other attributes, are used by Yelp’s ranking algorithm to rank businesses in a neighborhood. Due to Yelp’s substantial popularity and its high impact on local businesses’ success, understanding the bias of its algorithms is crucial. This data-driven study, for the first time, investigates the bias of Yelp’s business ranking and review recommendation system. We examine three hypotheses to assess if Yelp’s recommendation software shows bias against reviews of less established users with fewer friends and reviews and if Yelp’s business ranking algorithm shows bias against restaurants located in specific neighborhoods, particularly in hotspot regions, with specific demographic compositions. Our findings show that reviews of less-established users are disproportionately categorized as not-recommended. We also find a positive association between restaurants’ location in hotspot regions and their average exposure. Furthermore, we observed some cases of severe disparity bias in cities where the hotspots are in neighborhoods with less demographic diversity or higher affluence and education levels.
- Open Access
- 22 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 3762-3771
Causal Insights into Parler's Content Moderation Shift: Effects on Toxicity and Factuality
- 22 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 3762-3771
Open AccessCausal Insights into Parler's Content Moderation Shift: Effects on Toxicity and Factuality
- Open Access
FairDeFace: Evaluating the Fairness and Adversarial Robustness of Face Obfuscation Methods
The lack of a common platform and benchmark datasets for evaluating face obfuscation methods has been a challenge, with every method being tested using arbitrary experiments, datasets, and metrics. While prior work has demonstrated that face recognition systems exhibit bias against some demographic groups, there exists a substantial gap in our understanding regarding the fairness of face obfuscation methods. Providing fair face obfuscation methods can ensure equitable protection across diverse demographic groups, especially since they can be used to preserve the privacy of vulnerable populations. To address these gaps, this paper introduces a comprehensive framework, named FairDeFace, designed to assess the adversarial robustness and fairness of face obfuscation methods. The framework introduces a set of modules encompassing data benchmarks, face detection and recognition algorithms, adversarial models, utility detection models, and fairness metrics. FairDeFace serves as a versatile platform where any face obfuscation method can be integrated, allowing for rigorous testing and comparison with other state-of-the-art methods. In its current implementation, FairDeFace incorporates 6 attacks, and several privacy, utility and fairness metrics. Using FairDeFace, and by conducting more than 500 experiments, we evaluated and compared the adversarial robustness of seven face obfuscation methods. This extensive analysis led to many interesting findings both in terms of the degree of robustness of existing methods and their biases against some gender or racial groups. FairDeFace also uses visualization of focused areas for both obfuscation and verification attacks to show not only which areas are mostly changed in the obfuscation process for some demographics, but also why they failed through focus area comparison of obfuscation and verification.
Open AccessFairDeFace: Evaluating the Fairness and Adversarial Robustness of Face Obfuscation Methods
The lack of a common platform and benchmark datasets for evaluating face obfuscation methods has been a challenge, with every method being tested using arbitrary experiments, datasets, and metrics. While prior work has demonstrated that face recognition systems exhibit bias against some demographic groups, there exists a substantial gap in our understanding regarding the fairness of face obfuscation methods. Providing fair face obfuscation methods can ensure equitable protection across diverse demographic groups, especially since they can be used to preserve the privacy of vulnerable populations. To address these gaps, this paper introduces a comprehensive framework, named FairDeFace, designed to assess the adversarial robustness and fairness of face obfuscation methods. The framework introduces a set of modules encompassing data benchmarks, face detection and recognition algorithms, adversarial models, utility detection models, and fairness metrics. FairDeFace serves as a versatile platform where any face obfuscation method can be integrated, allowing for rigorous testing and comparison with other state-of-the-art methods. In its current implementation, FairDeFace incorporates 6 attacks, and several privacy, utility and fairness metrics. Using FairDeFace, and by conducting more than 500 experiments, we evaluated and compared the adversarial robustness of seven face obfuscation methods. This extensive analysis led to many interesting findings both in terms of the degree of robustness of existing methods and their biases against some gender or racial groups. FairDeFace also uses visualization of focused areas for both obfuscation and verification attacks to show not only which areas are mostly changed in the obfuscation process for some demographics, but also why they failed through focus area comparison of obfuscation and verification.
- Open Access
Attacks against Abstractive Text Summarization Models through Lead Bias and Influence Functions
Large Language Models have introduced novel opportunities for text comprehension and generation. Yet, they are vulnerable to adversarial perturbations and data poisoning attacks, particularly in tasks like text classification and translation. However, the adversarial robustness of abstractive text summarization models remains less explored. In this work, we unveil a novel approach by exploiting the inherent lead bias in summarization models, to perform adversarial perturbations. Furthermore, we introduce an innovative application of influence functions, to execute data poisoning, which compromises the model's integrity. This approach not only shows a skew in the models behavior to produce desired outcomes but also shows a new behavioral change, where models under attack tend to generate extractive summaries rather than abstractive summaries.
Open AccessAttacks against Abstractive Text Summarization Models through Lead Bias and Influence Functions
Large Language Models have introduced novel opportunities for text comprehension and generation. Yet, they are vulnerable to adversarial perturbations and data poisoning attacks, particularly in tasks like text classification and translation. However, the adversarial robustness of abstractive text summarization models remains less explored. In this work, we unveil a novel approach by exploiting the inherent lead bias in summarization models, to perform adversarial perturbations. Furthermore, we introduce an innovative application of influence functions, to execute data poisoning, which compromises the model's integrity. This approach not only shows a skew in the models behavior to produce desired outcomes but also shows a new behavioral change, where models under attack tend to generate extractive summaries rather than abstractive summaries.
- Open Access
From Chatbots to PhishBots? -- Preventing Phishing scams created using ChatGPT, Google Bard and Claude
The advanced capabilities of Large Language Models (LLMs) have made them invaluable across various applications, from conversational agents and content creation to data analysis, research, and innovation. However, their effectiveness and accessibility also render them susceptible to abuse for generating malicious content, including phishing attacks. This study explores the potential of using four popular commercially available LLMs, i.e., ChatGPT (GPT 3.5 Turbo), GPT 4, Claude, and Bard, to generate functional phishing attacks using a series of malicious prompts. We discover that these LLMs can generate both phishing websites and emails that can convincingly imitate well-known brands and also deploy a range of evasive tactics that are used to elude detection mechanisms employed by anti-phishing systems. These attacks can be generated using unmodified or "vanilla" versions of these LLMs without requiring any prior adversarial exploits such as jailbreaking. We evaluate the performance of the LLMs towards generating these attacks and find that they can also be utilized to create malicious prompts that, in turn, can be fed back to the model to generate phishing scams - thus massively reducing the prompt-engineering effort required by attackers to scale these threats. As a countermeasure, we build a BERT-based automated detection tool that can be used for the early detection of malicious prompts to prevent LLMs from generating phishing content. Our model is transferable across all four commercial LLMs, attaining an average accuracy of 96% for phishing website prompts and 94% for phishing email prompts. We also disclose the vulnerabilities to the concerned LLMs, with Google acknowledging it as a severe issue. Our detection model is available for use at Hugging Face, as well as a ChatGPT Actions plugin.
Open AccessFrom Chatbots to PhishBots? -- Preventing Phishing scams created using ChatGPT, Google Bard and Claude
The advanced capabilities of Large Language Models (LLMs) have made them invaluable across various applications, from conversational agents and content creation to data analysis, research, and innovation. However, their effectiveness and accessibility also render them susceptible to abuse for generating malicious content, including phishing attacks. This study explores the potential of using four popular commercially available LLMs, i.e., ChatGPT (GPT 3.5 Turbo), GPT 4, Claude, and Bard, to generate functional phishing attacks using a series of malicious prompts. We discover that these LLMs can generate both phishing websites and emails that can convincingly imitate well-known brands and also deploy a range of evasive tactics that are used to elude detection mechanisms employed by anti-phishing systems. These attacks can be generated using unmodified or "vanilla" versions of these LLMs without requiring any prior adversarial exploits such as jailbreaking. We evaluate the performance of the LLMs towards generating these attacks and find that they can also be utilized to create malicious prompts that, in turn, can be fed back to the model to generate phishing scams - thus massively reducing the prompt-engineering effort required by attackers to scale these threats. As a countermeasure, we build a BERT-based automated detection tool that can be used for the early detection of malicious prompts to prevent LLMs from generating phishing content. Our model is transferable across all four commercial LLMs, attaining an average accuracy of 96% for phishing website prompts and 94% for phishing email prompts. We also disclose the vulnerabilities to the concerned LLMs, with Google acknowledging it as a severe issue. Our detection model is available for use at Hugging Face, as well as a ChatGPT Actions plugin.
- Open Access
A Large-Scale Analysis of Phishing Websites Hosted on Free Web Hosting Domains
Free Website Building services (FWBs) provide individuals with a cost-effective and convenient way to create a website without requiring advanced technical knowledge or coding skills. However, malicious actors often abuse these services to host phishing websites. In this work, we propose FreePhish, a scalable framework to continuously identify phishing websites that are created using FWBs. Using FreePhish, we were able to detect and characterize more than 31.4K phishing URLs that were created using 17 unique free website builder services and shared on Twitter and Facebook over a period of six months. We find that FWBs provide attackers with several features that make it easier to create and maintain phishing websites at scale while simultaneously evading anti-phishing countermeasures. Our study indicates that anti-phishing blocklists and browser protection tools have significantly lower coverage and high detection time against FWB phishing attacks when compared to regular (self-hosted) phishing websites. While our prompt disclosure of these attacks helped some FWBs to remove these attacks, we found several others who were slow at removal or did not remove them outright, with the same also being true for Twitter and Facebook. Finally, we also provide FreePhish as a free Chromium web extension that can be utilized to prevent end-users from accessing potential FWB-based phishing attacks.
Open AccessA Large-Scale Analysis of Phishing Websites Hosted on Free Web Hosting Domains
Free Website Building services (FWBs) provide individuals with a cost-effective and convenient way to create a website without requiring advanced technical knowledge or coding skills. However, malicious actors often abuse these services to host phishing websites. In this work, we propose FreePhish, a scalable framework to continuously identify phishing websites that are created using FWBs. Using FreePhish, we were able to detect and characterize more than 31.4K phishing URLs that were created using 17 unique free website builder services and shared on Twitter and Facebook over a period of six months. We find that FWBs provide attackers with several features that make it easier to create and maintain phishing websites at scale while simultaneously evading anti-phishing countermeasures. Our study indicates that anti-phishing blocklists and browser protection tools have significantly lower coverage and high detection time against FWB phishing attacks when compared to regular (self-hosted) phishing websites. While our prompt disclosure of these attacks helped some FWBs to remove these attacks, we found several others who were slow at removal or did not remove them outright, with the same also being true for Twitter and Facebook. Finally, we also provide FreePhish as a free Chromium web extension that can be utilized to prevent end-users from accessing potential FWB-based phishing attacks.