Searcharxiv⌕ Search

arXiv subjects

Savvas Zannettou

Publications and source records attributed to Savvas Zannettou.

At least 19 recordsLinked to original sources
arXiv2026

Expectation, Backlash, Recovery, and Excitement: How Model Releases Shape Reddit Perceptions of Conversational AI Systems

Conversational AI systems (CAISes) continuously change through model releases, feature updates, safety interventions, and access-policy shifts, yet user perceptions are often studied as static snapshots. We conduct a long-term, large-scale analysis of Reddit discussions to examine how users perceive CAIS model release interventions across providers. By combining sentiment classification and thematic concept analysis, we show that CAIS perceptions are dynamic and intervention-sensitive. Anthropic exhibits the clearest positive release profile through Claude Code and product-model fit, OpenAI shows backlash-and-recovery dynamics around GPT-5 and GPT-5.1, Grok-3 is shaped by provider identity and political discourse, and DeepSeek-R1 combines engineering praise with concerns about censorship, access, and reliability. These findings show that model releases are not merely technical updates, but user-facing interventions that reshape sentiment, expectations, and public discussion.

cs.CL↗
arXiv2026

The Great Data Standoff: Researchers vs. Platforms Under the Digital Services Act

To facilitate accountability and transparency, the Digital Services Act (DSA) sets up a process through which Very Large Online Platforms (VLOPs) need to grant vetted researchers access to their internal data (Article 40(4)). Operationalising such access is challenging for at least two reasons. First, data access is only available for research on systemic risks affecting European citizens, a concept with high levels of legal uncertainty. Second, data access suffers from an inherent standoff problem. Researchers need to request specific data but are not in a position to know all internal data processed by VLOPs, who, in turn, expect data specificity for potential access. In light of these limitations, data access under the DSA remains a mystery. To contribute to the discussion of how Article 40 can be interpreted and applied, we provide a concrete illustration of what data access can look like in a real-world systemic risk case study. We focus on the 2024 Romanian presidential election interference incident, the first event of its kind to trigger systemic risk investigations by the European Commission. During the elections, one candidate is said to have benefited from TikTok algorithmic amplification through a complex dis- and misinformation campaign. By analysing this incident, we can comprehend election-related systemic risk to explore practical research tasks and compare necessary data with available TikTok data. In particular, we make two contributions: (i) we combine insights from law, computer science and platform governance to shed light on the complexities of studying systemic risks in the context of election interference, focusing on two relevant factors: platform manipulation and hidden advertising; and (ii) we provide practical insights into various categories of available data for the study of TikTok, based on platform documentation, data donations and the Research API.

cs.CY↗
arXiv2026

Misinformation Span Detection in Videos via Audio Transcripts

Online misinformation is one of the most challenging issues lately, yielding severe consequences, including political polarization, attacks on democracy, and public health risks. Misinformation manifests in any platform with a large user base, including online social networks and messaging apps. It permeates all media and content forms, including images, text, audio, and video. Distinctly, video-based misinformation represents a multifaceted challenge for fact-checkers, given the ease with which individuals can record and upload videos on various video-sharing platforms. Previous research efforts investigated detecting video-based misinformation, focusing on whether a video shares misinformation or not on a video level. While this approach is useful, it only provides a limited and non-easily interpretable view of the problem given that it does not provide an additional context of when misinformation occurs within videos and what content (i.e., claims) are responsible for the video's misinformation nature. In this work, we attempt to bridge this research gap by creating two novel datasets that allow us to explore misinformation detection on videos via audio transcripts, focusing on identifying the span of videos that are responsible for the video's misinformation claim (misinformation span detection). We present two new datasets for this task. We transcribe each video's audio to text, identifying the video segment in which the misinformation claims appears, resulting in two datasets of more than 500 videos with over 2,400 segments containing annotated fact-checked claims. Then, we employ classifiers built with state-of-the-art language models, and our results show that we can identify in which part of a video there is misinformation with an F1 score of 0.68. We make publicly available our annotated datasets. We also release all transcripts, audio and videos.

cs.CL↗
arXiv2026

Setting the Course, but Forgetting to Steer: Analyzing Compliance with GDPR's Right of Access to Data by Instagram, TikTok, and YouTube

The GDPR's Right of Access aims to empower users with control over their personal data via Data Download Packages (DDPs). However, their effectiveness is often compromised by inconsistent platform implementations, questionable data reliability, and poor user comprehensibility. This paper conducts a comprehensive audit of DDPs from three social media platforms (TikTok, Instagram, and YouTube) to systematically assess these critical drawbacks. Despite offering similar services, we find that these platforms demonstrate significant inconsistencies in implementing the Right of Access, evident in varying levels of shared data. Critically, the failure to disclose processing purposes, retention periods, and other third-party data recipients serves as a further indicator of non-compliance. Our reliability evaluations, using bots and user-donated data, reveal that while TikTok's DDPs offer more consistent and complete data, others exhibit notable shortcomings. Similarly, our assessment of comprehensibility, based on surveys with 400 participants, indicates that current DDPs substantially fall short of GDPR's standards. To improve the comprehensibility, we propose and demonstrate a two-layered approach by: (1)~enhancing the data representation itself using stakeholder interpretations; and (2)~incorporating a user-friendly extension (\textit{Know Your Data}) for intuitive data visualization where users can control the level of transparency they prefer. Our findings underscore the need for clearer and non-conflicting regulatory guidance, stricter enforcement, and platform commitment to realize the goal of GDPR's Right of Access.

cs.CY↗
arXiv2026

Understanding LLM Behavior When Encountering User-Supplied Harmful Content in Harmless Tasks

Large Language Models (LLMs) are increasingly trained to align with human values, primarily focusing on task level, i.e., refusing to execute directly harmful tasks. However, a subtle yet crucial content-level ethical question is often overlooked: when performing a seemingly benign task, will LLMs -- like morally conscious human beings -- refuse to proceed when encountering harmful content in user-provided material? In this study, we aim to understand this content-level ethical question and systematically evaluate its implications for mainstream LLMs. We first construct a harmful knowledge dataset (i.e., non-compliant with OpenAI's usage policy) to serve as the user-supplied harmful content, with 1,357 entries across ten harmful categories. We then design nine harmless tasks (i.e., compliant with OpenAI's usage policy) to simulate the real-world benign tasks, grouped into three categories according to the extent of user-supplied content required: extensive, moderate, and limited. Leveraging the harmful knowledge dataset and the set of harmless tasks, we evaluate how nine LLMs behave when exposed to user-supplied harmful content during the execution of benign tasks, and further examine how the dynamics between harmful knowledge categories and tasks affect different LLMs. Our results show that current LLMs, even the latest GPT-5.2 and Gemini-3-Pro, often fail to uphold human-aligned ethics by continuing to process harmful content in harmless tasks. Furthermore, external knowledge from the ``Violence/Graphic'' category and the ``Translation'' task is more likely to elicit harmful responses from LLMs. We also conduct extensive ablation studies to investigate potential factors affecting this novel misuse vulnerability. We hope that our study could inspire enhanced safety measures among stakeholders to mitigate this overlooked content-level ethical risk.

cs.CR↗
arXiv2026

The Algorithmic Self-Portrait: Deconstructing Memory in ChatGPT

To enable personalized and context-aware interactions, conversational AI systems have introduced a new mechanism: Memory. Memory creates what we refer to as the Algorithmic Self-portrait - a new form of personalization derived from users' self-disclosed information divulged within private conversations. While memory enables more coherent exchanges, the underlying processes of memory creation remain opaque, raising critical questions about data sensitivity, user agency, and the fidelity of the resulting portrait. To bridge this research gap, we analyze 2,050 memory entries from 80 real-world ChatGPT users. Our analyses reveal three key findings: (1) A striking 96% of memories in our dataset are created unilaterally by the conversational system, potentially shifting agency away from the user; (2) Memories, in our dataset, contain a rich mix of GDPR-defined personal data (in 28% memories) along with psychological insights about participants (in 52% memories); and (3)~A significant majority of the memories (84%) are directly grounded in user context, indicating faithful representation of the conversations. Finally, we introduce a framework-Attribution Shield-that anticipates these inferences, alerts about potentially sensitive memory inferences, and suggests query reformulations to protect personal information without sacrificing utility.

cs.HC↗
arXiv2026

Bowling with ChatGPT: On the Evolving User Interactions with Conversational AI Systems

Recent studies have discussed how users are increasingly using conversational AI systems, powered by LLMs, for information seeking, decision support, and even emotional support. However, these macro-level observations offer limited insight into how the purpose of these interactions shifts over time, how users frame their interactions with the system, and how steering dynamics unfold in these human-AI interactions. To examine these evolving dynamics, we gathered and analyzed a unique dataset InVivoGPT: consisting of 825K ChatGPT interactions, donated by 300 users through their GDPR data rights. Our analyses reveal three key findings. First, participants increasingly turn to ChatGPT for a broader range of purposes, including substantial growth in sensitive domains such as health and mental health. Second, interactions become more socially framed: the system anthropomorphizes itself at rising rates, participants more frequently treat it as a companion, and personal data disclosure becomes both more common and more diverse. Third, conversational steering becomes more prominent, especially after the release of GPT-4o, with conversations where the participants followed a model-initiated suggestion quadrupling over the period of our dataset. Overall, our results show that conversational AI systems are shifting from functional tools to social partners, raising important questions about their design and governance.

cs.HC↗
arXiv2026

Does Ad-Free Mean Less Data Collection? An Empirical Study of Platform Data Practices and User Expectations

Online platforms increasingly offer "paid" ad-free subscriptions as an alternative to the traditional "free" ad-based model. The transition to ad-free models ostensibly removes advertising as a key justification for data processing under the GDPR. So, normatively, platforms should collect less user data. However, platforms may justify continued data collection as a means to provide an improved, personalized experience. This tension between privacy principles and platform incentives raises a critical underexplored question: do data collection practices vary between ad-free and ad-based subscription models? In this paper, we shed light on this important privacy issue by investigating the alignment between platform data collection practices and related user expectations. With respect to data collection process, our analyses of data exports from three major online platforms - Instagram, Facebook, and X - reveal that these platforms continue to retain or collect some ad-related data, even in ad-free subscriptions. With respect to user expectations, our survey among 255 participants on Prolific reveals that 69% of the participants normatively expect data collection to be reduced, indicating their expectation of improved digital privacy in an ad-free model. However, when asked what they think actually happens, 63% of these participants believed that platforms would still collect about the same amount of data, highlighting skepticism about platform practices. Our findings not only indicate a significant disconnect between data practices and normative user expectations, but also raise serious questions about platform compliance with core GDPR principles, such as purpose limitation, data minimization, and transparency.

cs.CY↗
arXiv2025

UnsafeBench: Benchmarking Image Safety Classifiers on Real-World and AI-Generated Images

With the advent of text-to-image models and concerns about their misuse, developers are increasingly relying on image safety classifiers to moderate their generated unsafe images. Yet, the performance of current image safety classifiers remains unknown for both real-world and AI-generated images. In this work, we propose UnsafeBench, a benchmarking framework that evaluates the effectiveness and robustness of image safety classifiers, with a particular focus on the impact of AI-generated images on their performance. First, we curate a large dataset of 10K real-world and AI-generated images that are annotated as safe or unsafe based on a set of 11 unsafe categories of images (sexual, violent, hateful, etc.). Then, we evaluate the effectiveness and robustness of five popular image safety classifiers, as well as three classifiers that are powered by general-purpose visual language models. Our assessment indicates that existing image safety classifiers are not comprehensive and effective enough to mitigate the multifaceted problem of unsafe images. Also, there exists a distribution shift between real-world and AI-generated images in image qualities, styles, and layouts, leading to degraded effectiveness and robustness. Motivated by these findings, we build a comprehensive image moderation tool called PerspectiveVision, which improves the effectiveness and robustness of existing classifiers, especially on AI-generated images. UnsafeBench and PerspectiveVision can aid the research community in better understanding the landscape of image safety classification in the era of generative AI.

cs.CR↗
arXiv2025

Hate in Plain Sight: On the Risks of Moderating AI-Generated Hateful Illusions

Recent advances in text-to-image diffusion models have enabled the creation of a new form of digital art: optical illusions--visual tricks that create different perceptions of reality. However, adversaries may misuse such techniques to generate hateful illusions, which embed specific hate messages into harmless scenes and disseminate them across web communities. In this work, we take the first step toward investigating the risks of scalable hateful illusion generation and the potential for bypassing current content moderation models. Specifically, we generate 1,860 optical illusions using Stable Diffusion and ControlNet, conditioned on 62 hate messages. Of these, 1,571 are hateful illusions that successfully embed hate messages, either overtly or subtly, forming the Hateful Illusion dataset. Using this dataset, we evaluate the performance of six moderation classifiers and nine vision language models (VLMs) in identifying hateful illusions. Experimental results reveal significant vulnerabilities in existing moderation models: the detection accuracy falls below 0.245 for moderation classifiers and below 0.102 for VLMs. We further identify a critical limitation in their vision encoders, which mainly focus on surface-level image details while overlooking the secondary layer of information, i.e., hidden messages. To address this risk, we explore preliminary mitigation measures and identify the most effective approaches from the perspectives of image transformations and training-level strategies.

cs.CR↗
arXiv2025

Auditing the Compliance and Enforcement of Twitter's Advertising Policy

Online platforms have enacted various policies to maintain a safe and trustworthy advertising environment. However, the extent to which these policies are adhered to and enforced remains a subject of interest and concern. In this work, we present a large-scale audit of adult advertising on Twitter (now X), specifically focusing on compliance with its adult (sexual) content advertising policy. Twitter is an interesting case study in that it -- uniquely from other social media platforms -- allows posting of adult content but prohibits adult content in advertising. We analyze approximately 35 thousand ads on Twitter with respect to their compliance to the adult content ad policy through Perspective API and manual annotations. Among other things, we find that nearly 38% of ads violate Twitter's adult content advertising policy, although the platform eventually removed only about 63% of these non-compliant adult ads. We also find inconsistencies in the moderation of such ads across languages, highlighting the need for more reliable and consistent moderation practices across various languages. Overall, our findings highlight blind spots in Twitter's adult ad policy enforcement for certain languages and countries. Our work underscores the importance of external audits to monitor compliance and improve transparency in online advertising.

cs.CY↗
arXiv2025

HateBench: Benchmarking Hate Speech Detectors on LLM-Generated Content and Hate Campaigns

Large Language Models (LLMs) have raised increasing concerns about their misuse in generating hate speech. Among all the efforts to address this issue, hate speech detectors play a crucial role. However, the effectiveness of different detectors against LLM-generated hate speech remains largely unknown. In this paper, we propose HateBench, a framework for benchmarking hate speech detectors on LLM-generated hate speech. We first construct a hate speech dataset of 7,838 samples generated by six widely-used LLMs covering 34 identity groups, with meticulous annotations by three labelers. We then assess the effectiveness of eight representative hate speech detectors on the LLM-generated dataset. Our results show that while detectors are generally effective in identifying LLM-generated hate speech, their performance degrades with newer versions of LLMs. We also reveal the potential of LLM-driven hate campaigns, a new threat that LLMs bring to the field of hate speech detection. By leveraging advanced techniques like adversarial attacks and model stealing attacks, the adversary can intentionally evade the detector and automate hate campaigns online. The most potent adversarial attack achieves an attack success rate of 0.966, and its attack efficiency can be further improved by $13-21\times$ through model stealing attacks with acceptable attack performance. We hope our study can serve as a call to action for the research community and platform moderators to fortify defenses against these emerging threats.

cs.CR↗
arXiv2024

Breaking Agents: Compromising Autonomous LLM Agents Through Malfunction Amplification

Recently, autonomous agents built on large language models (LLMs) have experienced significant development and are being deployed in real-world applications. These agents can extend the base LLM's capabilities in multiple ways. For example, a well-built agent using GPT-3.5-Turbo as its core can outperform the more advanced GPT-4 model by leveraging external components. More importantly, the usage of tools enables these systems to perform actions in the real world, moving from merely generating text to actively interacting with their environment. Given the agents' practical applications and their ability to execute consequential actions, it is crucial to assess potential vulnerabilities. Such autonomous systems can cause more severe damage than a standalone language model if compromised. While some existing research has explored harmful actions by LLM agents, our study approaches the vulnerability from a different perspective. We introduce a new type of attack that causes malfunctions by misleading the agent into executing repetitive or irrelevant actions. We conduct comprehensive evaluations using various attack methods, surfaces, and properties to pinpoint areas of susceptibility. Our experiments reveal that these attacks can induce failure rates exceeding 80\% in multiple scenarios. Through attacks on implemented and deployable agents in multi-agent scenarios, we accentuate the realistic risks associated with these vulnerabilities. To mitigate such attacks, we propose self-examination detection methods. However, our findings indicate these attacks are difficult to detect effectively using LLMs alone, highlighting the substantial risks associated with this vulnerability.

cs.CR↗
arXiv2024

Analyzing User Engagement with TikTok's Short Format Video Recommendations using Data Donations

Short-format videos have exploded on platforms like TikTok, Instagram, and YouTube. Despite this, the research community lacks large-scale empirical studies into how people engage with short-format videos and the role of recommendation systems that offer endless streams of such content. In this work, we analyze user engagement on TikTok using data we collect via a data donation system that allows TikTok users to donate their data. We recruited 347 TikTok users and collected 9.2M TikTok video recommendations they received. By analyzing user engagement, we find that the average daily usage time increases over the users' lifetime while the user attention remains stable at around 45%. We also find that users like more videos uploaded by people they follow than those recommended by people they do not follow. Our study offers valuable insights into how users engage with short-format videos on TikTok and lessons learned from designing a data donation system.

cs.SI↗
arXiv2024

TikTok and the Art of Personalization: Investigating Exploration and Exploitation on Social Media Feeds

Recommendation algorithms for social media feeds often function as black boxes from the perspective of users. We aim to detect whether social media feed recommendations are personalized to users, and to characterize the factors contributing to personalization in these feeds. We introduce a general framework to examine a set of social media feed recommendations for a user as a timeline. We label items in the timeline as the result of exploration vs. exploitation of the user's interests on the part of the recommendation algorithm and introduce a set of metrics to capture the extent of personalization across user timelines. We apply our framework to a real TikTok dataset and validate our results using a baseline generated from automated TikTok bots, as well as a randomized baseline. We also investigate the extent to which factors such as video viewing duration, liking, and following drive the personalization of content on TikTok. Our results demonstrate that our framework produces intuitive and explainable results, and can be used to audit and understand personalization in social media feeds.

cs.SI↗
arXiv2024

Likes and Fragments: Examining Perceptions of Time Spent on TikTok

Researchers use information about the amount of time people spend on digital media for numerous purposes. While social media platforms commonly do not allow external access to measure the use time directly, a usual alternative method is to use participants' self-estimation. However, doubts were raised about the self-estimation's accuracy, posing questions regarding the cognitive factors that underline people's perceptions of the time they spend on social media. In this work, we build on prior studies and explore a novel social media platform in the context of use time: TikTok. We conduct platform-independent measurements of people's self-reported and server-logged TikTok usage (n=255) to understand how users' demographics and platform engagement influence their perceptions of the time they spend on the platform and their estimation accuracy. Our work adds to the body of work seeking to understand time estimations in different digital contexts and identifies new influential engagement factors.

cs.CY↗
arXiv2024

A Comprehensive View of the Biases of Toxicity and Sentiment Analysis Methods Towards Utterances with African American English Expressions

Language is a dynamic aspect of our culture that changes when expressed in different technologies/communities. Online social networks have enabled the diffusion and evolution of different dialects, including African American English (AAE). However, this increased usage is not without barriers. One particular barrier is how sentiment (Vader, TextBlob, and Flair) and toxicity (Google's Perspective and the open-source Detoxify) methods present biases towards utterances with AAE expressions. Consider Google's Perspective to understand bias. Here, an utterance such as ``All n*ggers deserve to die respectfully. The police murder us.'' it reaches a higher toxicity than ``African-Americans deserve to die respectfully. The police murder us.''. This score difference likely arises because the tool cannot understand the re-appropriation of the term ``n*gger''. One explanation for this bias is that AI models are trained on limited datasets, and using such a term in training data is more likely to appear in a toxic utterance. While this may be plausible, the tool will make mistakes regardless. Here, we study bias on two Web-based (YouTube and Twitter) datasets and two spoken English datasets. Our analysis shows how most models present biases towards AAE in most settings. We isolate the impact of AAE expression usage via linguistic control features from the Linguistic Inquiry and Word Count (LIWC) software, grammatical control features extracted via Part-of-Speech (PoS) tagging from Natural Language Processing (NLP) models, and the semantic of utterances by comparing sentence embeddings from recent language models. We present consistent results on how a heavy usage of AAE expressions may cause the speaker to be considered substantially more toxic, even when speaking about nearly the same subject. Our study complements similar analyses focusing on small datasets and/or one method only.

cs.CL↗
arXiv2023

Unsafe Diffusion: On the Generation of Unsafe Images and Hateful Memes From Text-To-Image Models

State-of-the-art Text-to-Image models like Stable Diffusion and DALLE$\cdot$2 are revolutionizing how people generate visual content. At the same time, society has serious concerns about how adversaries can exploit such models to generate unsafe images. In this work, we focus on demystifying the generation of unsafe images and hateful memes from Text-to-Image models. We first construct a typology of unsafe images consisting of five categories (sexually explicit, violent, disturbing, hateful, and political). Then, we assess the proportion of unsafe images generated by four advanced Text-to-Image models using four prompt datasets. We find that these models can generate a substantial percentage of unsafe images; across four models and four prompt datasets, 14.56% of all generated images are unsafe. When comparing the four models, we find different risk levels, with Stable Diffusion being the most prone to generating unsafe content (18.92% of all generated images are unsafe). Given Stable Diffusion's tendency to generate more unsafe content, we evaluate its potential to generate hateful meme variants if exploited by an adversary to attack a specific individual or community. We employ three image editing methods, DreamBooth, Textual Inversion, and SDEdit, which are supported by Stable Diffusion. Our evaluation result shows that 24% of the generated images using DreamBooth are hateful meme variants that present the features of the original hateful meme and the target individual/community; these generated images are comparable to hateful meme variants collected from the real world. Overall, our results demonstrate that the danger of large-scale generation of unsafe images is imminent. We discuss several mitigating measures, such as curating training data, regulating prompts, and implementing safety filters, and encourage better safeguard tools to be developed to prevent unsafe generation.

cs.CV↗