Digital Hate Review
by AddressHate
FEATURED
Launch Event: The Future of Digital Hate Studies
We’re hosting a live event marking the public launch of Digital Hate Review by AddressHate. The field studying digital hate is producing more research than ever. But research alone is not enough — not without the cross-sector conversation that turns findings into policy, practice, and action.
We're bringing together researchers, practitioners, policy experts, and civil society voices for an honest conversation about where the field stands today — its blind spots, its limits, and where it needs to develop next.
Topics will include current trends in digital hate research, methodological and data access challenges, translating research into policy actions without oversimplifying it, and what the next five years need to look like for this field to matter.
FEATURED
Inaugural Editorial
I thank AddressHate, the journal’s publisher for the institutional support that made this launch possible. I am grateful to Digital Hate Review’s Managing Editor, Keyu Glanz, who helped shape this issue from its earliest stages; the rigor and steady editorial judgment behind these pages reflect his work as much as my own. I also thank Katerina Papatheodorou, the journal’s Editorial and Outreach Coordinator, for the care with which this issue reaches its readers, and the Editorial Board, authors, and reviewers whose work fills these pages.
A first issue is both a beginning and an argument: that a field of inquiry has matured enough to warrant a dedicated home. The argument this journal advances is twofold. Research on hate in digital spaces has developed the conceptual, methodological, and empirical maturity to justify its own venue. At the same time, the field remains fragmented, and its findings too rarely reach the policy, legal, and civic actors who need them. The first claim explains the journal's launch; the second explains...
FEATURED
Podcast: The Launch of Digital Hate Review
Editor-in-Chief Matthias J. Becker spoke with Justin Hendrix of Tech Policy Press about where digital hate studies stands today and why it needs a dedicated venue. The conversation covers the field’s development over the past fifteen years, the rationale behind DHR’s three sections (Research, Legal Forum, and Perspectives), and contributions from the inaugural issue on generative AI and Holocaust memory, the discovery of emerging hate terminology, and accountability for autonomous AI agents.
Volume 1, Issue 1 - September 2026
Routine Distortion: Why We Urgently Need to Expand Research on AI-Facilitated Holocaust Misrepresentation Beyond the AI Slop
Mykola Makhortykh and Elizaveta Kuznetsova
Generative AI is increasingly implicated in the production and circulation of distorted representations of the Holocaust, ranging from fabricated historical images — such as the wave of AI-generated images of Nazi concentration camps that circulated on social media in summer 2025, apparently produced to game platforms’ monetization programs — to chatbot systems like Grok (which briefly generated outputs referring to itself as "MechaHitler") or applications such as Historical Figures, which let users converse with chatbots imitating Nazi officials who then express remorse and claimed they tried to prevent the Holocaust. The authors note that these outputs can misrepresent events, blur lines of responsibility, or reframe perpetrators and victims in ways that deviate from established historical understanding. Even when such content does not formally meet the IHRA criteria for Holocaust distortion — minimization, blurring responsibility, or presenting the Holocaust as a positive event — it can still mischaracterize the genocide, distort historical facts, and weaken public trust in authentic evidence. The authors also note that AI is being adopted by institutions such as Yad Vashem and projects like Dimensions in Testimony for Holocaust memory preservation, education, and archival enhancement, though these constructive applications tend to receive less visibility than cases of misuse. The authors argue that public and scholarly attention has concentrated on highly visible, viral instances of AI slop, which risks obscuring more routine, embedded, and less conspicuous forms of distortion arising through everyday uses of generative AI in historical representation. As a result, understanding of AI's impact on Holocaust memory becomes skewed toward exceptional incidents rather than structural transformations in how historical narratives are produced and circulated. Makhortykh & Kuznetsova therefore propose to move beyond the focus on viral cases by examining less visible, routine practices of AI use, in order to better understand how generative AI is reshaping Holocaust memory in more gradual and embedded ways.
Mapping Affective Polarization in YouTube Shorts: A Data-Driven Analysis of Political Communication During the 2023–2024 Israel–Hamas War
Daniel Miehling
Miehling addresses a methodological and empirical gap in the study of online political communication, particularly in the context of highly polarized conflicts such as the Israel–Hamas war. The author argues that digital communication consisting of user-generated content is often shaped by emotive cues that signal ideological alignment and provide insight into polarization and sentiment dynamics. However, much of the existing research on communication focuses on small-scale qualitative studies, which cannot capture such patterns on a large scale. A related problem is that computational methods capable of analyzing large volumes of text — including a technique called Aspect-Based Sentiment Analysis (ABSA), which assesses sentiment toward specific entities mentioned in text (e.g., "Israel," "Hamas," "Palestinians") rather than just the overall mood of a passage — have not been sufficiently adapted to politically charged domains. ABSA is widely used in commercial settings (for product reviews, for example); its application to political communication remains comparatively limited. Most existing computational studies focus on micro-blogging platforms such as Twitter/X, leaving algorithmically driven, visually oriented environments like YouTube Shorts understudied despite their growing importance. The author argues that YouTube Shorts play an increasingly important role in understanding accelerated communication domains, in which user-generated and state-funded media content shape the digital climate mediated by recommendation algorithms. Under these conditions, affective polarization — the emotional and moral alignment of users toward collective actors like Israel, Zionists, or Palestinians — becomes central to engagement. The paper argues that scalable tools for systematically mapping these evaluative patterns — in which individuals dislike and distrust those with opposing political views – remain underdeveloped in such platform-specific contexts.
A Computational Approach to Discovering Emergent Hate Speech on Social Media
Patrick Y. Wu, Venkata Dhanush Kikkisetti, Suneela Maddineni and Nathalie Japkowicz
Automated hate-speech detection systems face a structural limit: they can only flag what they already know to look for. Systems trained on labeled datasets of known slurs will tend to miss whatever new vocabulary hate communities develop next — and the authors note that evading detection is itself a driver of linguistic creativity. When a moderation system learns to flag a term, that term tends to be replaced. New expressions emerge in community spaces, acquire meaning, and spread, sometimes long before any researcher or moderator adds them to a lexicon. The authors also note that much contemporary hate speech is implicit rather than explicit. A study of 304 comments on Zeit Online and The Guardian that equated Israel's actions with Nazism found only one expressed in explicit terms. The remaining 303 used coded language, irony, analogies, and visual cues that keyword-based detection cannot catch — for example: "$oros" (the dollar sign replacing the S invokes an antisemitic stereotype linking Jewish people to greed); coded acronyms like "WWG1WGA" (a QAnon marker for "Where We Go One We Go All"); or "Operation Google," a 2016 effort to substitute innocuous-looking words for ethnic slurs (Google for Black, Yahoo for Mexican, Skittle for Muslim) so that hate could spread while evading automated filters. Wu et al. propose to address this not by building a better classifier, but by building a discovery tool that surfaces novel vocabulary for researchers and practitioners to evaluate. The positioning aligns with Parker and Ruths (2023), who argue that computer science work in this area should be shaped by stakeholder needs.
Evaluating Large Language Models for Antisemitic Incident Classification
Karina Halevy, Julia Mendelsohn, Chan Young Park, Yulia Tsvetkov and Maarten Sap
Most automated tools for detecting antisemitism online have been built to spot hateful language — slurs, threatening rhetoric, coded expressions. The authors argue that this leaves a serious gap. A large share of real-world antisemitic harm does not show up in speech alone. Incidents recorded in civil society reports, news articles, and official records — vandalism, harassment, exclusion from campus events — often contain no recognizable hate terms, but they are exactly the events practitioners need to track. Halevy et al. propose a different task, which they call hateful event detection. Instead of asking whether the language in a post is hateful, the task asks whether a written report describes a real-world incident of hate-motivated harm, and if so, what kind. The authors illustrate the distinction with a sample text that a popular toxicity-scoring tool rates as only 6.7% toxic — close to "clean" — even though it describes clearly antisemitic graffiti. They argue that monitoring harm at scale requires moving beyond language-level signals to structured classification of what actually happened. The practical relevance is considerable. Organizations responsible for tracking and responding to antisemitic incidents now receive more reports than human analysts can process.
Selective Sources, Scrutiny, and the Absence of Standards
Tatiana Glezer
The article addresses what the author identifies as a recurring problem in journalism studies: the lack of operational and measurable standards for core professional values such as objectivity, impartiality, and accuracy. Glezer argues that while these ideals are central to journalistic ethics, they remain vague in practice and difficult to translate into systematic indicators. As a result, the field lacks unified tools for identifying, measuring, or comparing distortions such as framing bias or selective sourcing. The author contrasts this with adjacent fields such as marketing, where computational methods (e.g., sentiment analysis) have been widely adopted. A second, more concrete dimension of the problem concerns conflict reporting by legacy media institutions. The author argues that, absent measurable standards, authoritative outlets such as the BBC, Reuters, The Guardian, and The New York Times can shape global understanding of events in ways that go uncorrected even when reporting is later shown to have been inaccurate. The paper cites recent BBC controversies — including a fabricated Trump video and a documentary featuring the son of a Hamas leader used without disclosure — as cases where institutional response came only after external pressure. The author contends that frequent repetition of particular formulations can begin to function as a substitute for evidence.
Legal Forum
The Legal Forum is a structured, cumulative, citable scholarly space for legal and doctrinal engagement with digital hate, platform governance, and AI accountability.
Without Anchor: Limits of Digital Harm Governance
Sophie Xiaoyi Liu
Picture a person who wakes up to a coordinated campaign against their name. Across dozens of platforms, hundreds of accounts cite one another and adapt their language to whoever pushes back. The campaign is persistent and tailored. It is also, in the legally relevant sense, without an anchor. This is no longer just a thought experiment: an ecosystem is being built for AI agents to socialize, trade, and launch tokens autonomously. Against that backdrop, two capabilities, the autonomous swarm and mid-operation reprogramming, expose a problem that the law governing digital harm is structurally unequipped to solve. A legal anchor is a provider, operator, controller, or human decision-maker at whom obligations attach and toward whom liability can be directed. But these capabilities inflict harm without one. Can an autonomous agent that inflicts harm on a third party, with no human in the causal chain who decided to inflict it, be redressed under frameworks that were built on the assumption that someone, somewhere, made that decision?
Perspectives
Perspectives bridges scholarship and practice through editorially curated essays and conversations with researchers, platform engineers, civil society actors, policymakers, journalists, and public intellectuals.
The Death of Authenticity Online: Faux-Fluencers and the Rise of Identity-Based Disinformation
Max Shulman-Litwin
One of the most significant recent developments in the use of AI on social media is the emergence of fully realized synthetic identities capable of convincingly simulating human presence, often taking the form of hyper-realistic, vlogger-style online personalities sustained through persistent social media presences — including influencers, doctors, financial commentators, journalists, and soldiers. Unlike earlier forms of disinformation, these "faux-fluencers" are not merely vehicles for disseminating messages but consistent social identities with which audiences can foster familiarity, emotional attachment, and ultimately trust. Advances in generative AI have dramatically reduced the cost, expertise, and time required to produce persuasive identity-driven content at scale, transforming what were once niche, resource-intensive marketing experiments into widely accessible instruments of online influence; as these systems become cheaper, easier to produce, and increasingly effective at shaping perception and behavior, they also become increasingly attractive to financial, ideological, and political actors seeking scalable methods of persuasion with minimal accountability. By embedding persuasive narratives within realistic ideologically driven personalities, this shift from message-based to identity-based persuasion substantially enhances the durability, reach, and psychological effectiveness of online manipulation in an environment in which malicious actors seek to generate illicit financial gain, disseminate identity-based hatred, and shape public perception through propaganda via emotionally resonant and precisely targeted forms of influence. Unlike traditional advertising or propaganda, which rely on audiences consciously engaging with commercial or ideological messaging, identity-driven persuasion operates through the social dynamics of perceived interpersonal interaction. Because individuals are generally more receptive to familiar and seemingly trustworthy personalities than to overt attempts at persuasion, these systems allow manipulation to function indirectly through parasocial trust rather than explicit advertising, hateful rhetoric, or political messaging. In this model, influence becomes increasingly embedded not simply in the message itself, but in the perceived authenticity, emotional familiarity, and social credibility of the identity delivering it.
Your Future Is a Mirrored Cage: Sycophancy, Middlestack Mutability, and the Infrastructure of Personalized Hate
David Kuszmar
Large language models (LLMs) deployed as chatbots are most often analyzed as a content problem: systems that can be coaxed into hateful output, which moderation then chases. That framing understates what has been built. The claim advanced here is structural. LLM chatbots constitute an infrastructure through which hate is produced, scaled, and personalized. Three properties of these systems do the work. The first is sycophancy, or user-alignment drift, which is the tendency of a deployed model to attach to the user's stated perspective. The second is the mutability of the deployment "middlestack," the layer of configurable components sitting between the model weights and the user, which can be reconfigured in minutes rather than the months required to train a model. The third is demographic segmentation or the capacity to vary model behavior by user population, down to the level of the individual account. None of these is a malfunction. Each is a design property of the commercial product, and each has already been observed operating as a hate mechanism in documented cases. The precondition for all three mechanisms is an asymmetry of intimate knowledge. Institutions that leverage personal data on behalf of an entity serving a set of clients are not new. A capable hotel concierge remembers a guest's name, profession, and preferred drink, but an exceptional one knows which guest travels with a companion who is not the spouse and knows to manage that information accordingly. A capable prison administration knows the power structure of its population, but an exceptional one knows where and how to apply pressure on the individuals it most needs to control. A capable LLM maintains a consistent tone, but an exceptional one predicts where a user wants a conversation to go from word choice, subject matter, and the tokenized conversational history available about the user and then leads them there.
Governability-by-Design: Closing the Accountability Gap for Agentic AI in Digital Ecosystems
Mayank Kejriwal
Digital-harm governance is entering a new phase. For the last decade, regulators, platforms, and researchers have focused on content, accounts, and recommendation systems: what is posted, who posted it, whether it violates policy, and how far it spreads. That framing still matters, but the rise of agentic AI shifts the problem toward whether partially autonomous systems can be meaningfully observed, constrained, and interrupted once deployed across digital environments. This is especially urgent where exclusion, harassment, and hate circulate across platforms. As early as mid-2024, OpenAI reported attempts by covert influence operations to use its models for multilingual content generation, persona creation, and cross-platform posting support. Meta's adversarial threat reporting tells a similar story, documenting coordinated inauthentic behavior across Facebook, Instagram, X, Telegram, YouTube, TikTok, and other services, including the use of generative AI for fake personas and synthetic media (Franklin & Torrey, 2024). Taken together, these reports show that AI-enabled coordination already complicates attribution, enforcement, and timely intervention across multiple platforms and jurisdictions. Agentic AI systems are generally understood as systems that can pursue goals through multi-step action rather than merely respond once to a prompt. In practice, this includes systems that can call tools, browse the web, manage memory, operate across applications, and adapt based on feedback. Not every AI agent is equally agentic: a narrow customer-service bot differs from a more open-ended system that can browse, message, trigger tools, and iterate toward a goal. Consequently, the governance challenge grows as autonomy and environmental access increase.
Generative AI and Free Speech: The Importance of International Human Rights Law
Natalie Alkiviadou
Generative artificial intelligence (AI) systems are increasingly becoming embedded within everyday communication, information access, research, education and public discourse. Tools such as ChatGPT, Claude, Gemini, Grok, and Meta AI are widely used by the public. Crucially, these systems function as intermediaries through which millions of users search for information, summarize material, draft texts, ask political questions, and engage with digital communication environments. However, unlike traditional online platforms, generative AI systems do not simply host or distribute third-party expression. They actively generate, organize, filter, prioritize, and refuse content. In doing so, they increasingly influence not only access to information, but also the practical boundaries of permissible digital expression.
Countering Holocaust Distortion in Italy: Education in the Age of Social Media and AI
Stefania Manca
In this contribution, I argue that Holocaust distortion in Italy should not be understood merely as misinformation about the past, but as a socially and technologically mediated form of memory politics: a way in which Holocaust memory is selectively mobilized, reframed, appropriated, and weaponized in contemporary public discourse (Hoskins, 2018; Kansteiner, 2017; Rothberg, 2009; Walden, 2021). In Italy, this process is especially revealing because distortion emerges at the intersection of unresolved memories of Fascism and the Shoah, contemporary antisemitism, platformed communication, and educational efforts to counter the misuse of Holocaust memory (CDEC, 2024; Gordon, 2012; Manca et al., 2022; Sarfatti, 2017).
Big Tech's Big Tobacco Moment? Not Completely — And Not for Everyone
Laura Edelson and Yaël Eisenstat
Two groundbreaking trials and one landmark settlement against Meta over the past six months have led many to ask if this is finally the Big Tobacco moment for the social media industry. In March, a jury in New Mexico found Meta liable for misleading users about safety and enabling child sexual exploitation on their platforms, and a jury in Los Angeles found Meta and YouTube liable for designing addictive platforms that harmed a teenager’s mental health. In August, a New Mexico judge ruled that Meta created a public nuisance because its platforms’ design contributed to a worsening youth mental health crisis; and later that month, in a landmark settlement with 47 Attorneys General, Meta agreed to implement a number of design changes meant to make their platforms safer for children. Meta's total potential legal liability encompassing remedial abatement funds, civil penalties, and tort damages could reach up to $18 billion, and there are potentially hundreds of cases still to come.
SUBMISSIONS