Poojitha Thota

Hello!

I am a PhD Candidate in Computer Science and Engineering at the University of Texas at Arlington, advised by Dr. Shirin Nilizadeh. My research focuses on security and privacy in AI systems, particularly examining vulnerabilities of large language and vision-language models. I work on understanding how these systems can be exploited through adversarial attacks and data poisoning, and developing robust defenses against such threats. My work spans applications from text summarization to medical imaging, aiming to build more trustworthy and secure AI systems for real-world deployment.

I have been fortunate to intern twice at Google's Responsible AI team, working on improving the safety of Gemini models through multi-agent frameworks and adversarial prompt detection. My research has been recognized with several awards, including the Distinguished Paper Award at IEEE S&P 2024 and the Johns Hopkins Suchman Outstanding Doctoral Student Award at UTA. Before my PhD, I worked as a Graduate Engineer at Hyundai Mobis, developing parking application software. I received my Master's and Bachelor's degrees in Computer Science and Electronics & Communication, respectively. My transition to security and privacy research was driven by the increasing importance of building trustworthy AI systems that can withstand adversarial threats in real-world applications.

I'm always open to research collaborations and happy to chat about AI security and privacy. Feel free to drop me an email!

Research Highlights

	Learning from Censored Experiences: Social Media Discussions around Censorship Circumvention Technologies IEEE S&P 2025 Analyzed social media discussions about censorship circumvention technologies to understand user experiences and information sharing patterns around internet censorship. Elham Pourabbas Vafa, Mohit Singhal, Poojitha Thota, Sayak Saha Roy. PAPER
	Attacks against Abstractive Text Summarization Models through Lead Bias and Influence Functions EMNLP 2024 Investigated vulnerabilities in text summarization models through adversarial perturbations and data poisoning attacks. Exploited lead bias for adversarial perturbations and used influence functions for data poisoning attacks. Poojitha Thota, Shirin Nilizadeh PAPER CODE
	From Chatbots to Phishbots?: Phishing Scam Generation in Commercial Large Language Models IEEE S&P 2024 - Distinguished Paper Award Explored the capabilities of commercial LLMs to produce evasive phishing attacks by crafting malicious prompts. Designed a classifier for early detection of malicious prompts achieving 98% accuracy using fine-tuned RoBERTa. Roy, Sayak Saha, Poojitha Thota, Krishna Vamsi Naragam, Shirin Nilizadeh PAPER MODEL CODE
	Demonstration of an Adversarial Attack Against a Multimodal Vision Language Model for Pathology Imaging IEEE ISBI 2024 Successfully demonstrated adversarial attacks against PLIP, a vision-language model for pathology imaging, achieving 100% attack success rate using PGD. Poojitha Thota, Jai Prakash Veerla, Partha Sai Guttikonda, Mohammad S. Nasr, Jacob M. Luber, Shirin Nilizadeh PAPER CODE
	SoK: Content Moderation in Social Media, from Guidelines to Enforcement, and Research to Practice IEEE EuroS&P 2023 Systematized knowledge on content moderation practices across social media platforms, analyzing the gap between stated guidelines and actual enforcement mechanisms. Mohit Singhal, Chen Ling, Pujan Paudel, Poojitha Thota, Nihal Kumarswamy, Gianluca Stringhini, Shirin Nilizadeh PAPER
	Web scraping of covid-19 news stories to create datasets for sentiment and emotion analysis PETRA 2021 Analyzed emotion and sentiment in leaders' statements and news stories to understand their impact on COVID-19 cases. Built web scraper and applied BERT-based NLP techniques achieving 85.2% accuracy. Poojitha Thota, Elmasri Ramez PAPER

Hello!

Recent News

Research Highlights