This Dispatch contains an excerpt from an ISD expert report submitted in response to a Letter of Instruction from the Royal Commission on Antisemitism and Social Cohesion. The Royal Commission was set up following the December 2025 terrorist attack on the Jewish community in Bondi to understand the security failures that precipitated it and the wider environment of antisemitism in Australia.   

Antisemitism exists at the nexus of diverse extremist movements that target Jews based on a supremacist worldview. However, antisemitism is also present across the political mainstream through conspiracy theories, targeted hate towards Jewish people and social exclusion. ISD’s report documents the long tail of antisemitism and its impacts on Jewish people and wider society, from violent extremism to latent antisemitism online. 

The report provides evidence of the scale and nature of antisemitic hate online, the need for platform responsibility and good governance, approaches to enforcement and transparency, and different regulatory models around the world. Overall, it sets out the need for an approach to mitigating antisemitism on social media that is informed by the hybrid nature of the current harms landscape, future-proofed against evolving threats and underpinned by platform transparency.  

Key findings  

Overall findings

• Online antisemitism is characterised by a diffuse web of diverse ideological actors across mainstream and fringe spaces. This poses a major challenge to effective enforcement of platform policies. 
• Far-right, far-left and Islamist actors increasingly converge on antisemitic narratives, particularly conspiracy theories. Hamas’ attacks on 7 October 2023 were a major accelerant. 
• ISD has identified an increasing trend of antisemitic post-organisational online networks alongside established violent actors. These networks can exploit gaps in counter-extremism systems designed for group-based threats. 
• Iran-linked networks have employed a ‘violence-as-a-service’ model, recruiting financially motivated actors for attacks on Jewish communal targets. Hostile states also use social media to erode democratic resilience and cohesion at scale: Iranian diplomatic and official accounts generated approximately 900 million views and 22 million likes in the first 50 days of the April 2026 Iran war through sophisticated AI-assisted content strategies. 
• Antisemitic ideas are increasingly part of mainstream discourse. Following the Bondi Beach attack, false flag conspiracy theories and content calling for further violence circulated freely on mainstream platforms. 
• Platform recommender algorithms serve antisemitic content to users. ISD research shows that a 15-year-old boy with an interest in left-wing topics only needs to spend 90 minutes on TikTok before being recommended content aligned with Iran’s Islamic Revolutionary Guard Corps (IRGC). 

Australia-specific findings 

• Antisemitic content from accounts that identified themselves as located in Australia was driven by domestic and international events. AI-driven analysis of content produced on X (formerly known as Twitter), 4chan, Tumblr, extremist forums and extremist blogs between May 2025 and May 2026 found that the highest daily volumes correlated with the December 2025 Bondi Beach attack, the Iran–Israel war and the assassination of Charlie Kirk.  
• Fringe boards play an outsize role in the spread of antisemitic content. Of identified content, 82 percent was on 4chan’s /pol/ board; 16 percent on X. This data snapshot is almost certainly a significant underestimate of total Australian-derived antisemitic content
  

Case study: a snapshot of online antisemitism in Australia 

Methodology 

Analysts collected English-language antisemitic posts published on five platforms – 4chan, X, Tumblr, extremist forums and extremist blogs – between 26 May 2025 and 26 May 2026 from accounts which identified themselves as located in Australia. Platforms were selected for study on the basis of the available functionality to identify Australian users. After collecting posts containing terms from a list of over 1,000 identified as likely relevant to antisemitism, analysts prompted and refined a large language model (LLM) classifier to filter out noise, achieving 87 percent accuracy (see annex for full methodology). In total, 41,770 antisemitic posts were identified.  

This is likely to be a considerable underestimation of the volume of Australian-derived antisemitic content, given that users often do not identify their location or do so falsely. Furthermore, this does not include the significant volume of antisemitic content originating in other countries that is available to Australian audiences. Cross-platform comparisons of the volumes of antisemitic content are hampered by the lack of consistent access to data provided to third-party researchers, meaning that not all content could be captured.  

Identifying trends in antisemitic posts 

Figure 1: A graph showing the volume of antisemitic content per day over a one-year period from accounts identifying themselves as Australian. 

Peaks in volume clearly correlate with significant global events. The highest and second highest volumes of antisemitism identified on any single day were recorded on 22 and 13 June 2025 respectively. On 13 June, Israel attacked Iranian nuclear facilities and military bases. This prompted a 12-day war that the United States joined on 22 June. Users commonly celebrated Iranian attacks on Israel; they claimed that US intervention was evidence of deference to Israel and proof of a global Jewish cabal. This finding is consistent with ISD’s research showing an increase in online antisemitism during wars involving Iran, including a 68 percent increase in US-based antisemitic content in the week following the April 2026 war. Such spikes show how overseas events (particularly in the Middle East) threaten cohesion and community safety in Australia. These spikes further highlight the need for platforms to build and implement more effective event protocols. 

The third highest volume of antisemitic content appeared on the day of the Bondi Beach attack. Users claimed that Jews would do anything for sympathy, brushed off the severity of the attack by claiming that the victims were not Australian as they were Jewish, victim-blamed Jews for “doing it to yourself” and spread false flag conspiracy theories alleging Mossad was behind the attack. False flag claims – which are often antisemitic in nature – are increasingly common following atrocities, having grown by more than 1,100 percent on X over the past five years. Some posts, particularly on 4chan, called for more antisemitic violence, including “shoot[ing] a Jewish politician”. The Bondi Beach attack also prompted spikes in online antisemitism globally, including in Canada. These trends demonstrate how offline events prompt online antisemitism, which in turn can mobilise further offline violence. Policy responses must consider the cyclical relationship between online and offline environments. 

A final notable peak identified in the dataset occurred following the assassination of Charlie Kirk on 10 September 2025. Users shared claims that Jews or Israel were responsible, in line with wider conspiracy theories that Kirk’s recent change in position on Israel had prompted the assassination. This speaks to the transnationalism of online hate and the need for a coordinated intergovernmental response.  

Figure 2: A graph showing the volume of antisemitic content per week over a one-year period from accounts identifying themselves as Australian, disaggregated by platform.  

Figure 2 shows the volume of antisemitic content per week across the studied platforms. Of identified antisemitic content, 82 percent was located on 4chan’s /pol/ board, an online forum known to host far-right communities with little moderation. Of antisemitic content, 16 percent was identified on X. This data highlights the role of both very large platforms and small high-risk services in propagating antisemitic content. These platforms should be understood within a wider ecosystem of mainstream and fringe platforms, gaming servers and messenger services. Together they form a tech stack, with extremists exploiting different services for purposes including operational and strategic planning, mainstreaming, recruitment and in-group socialising. Policy responses must address larger and more fringe online platforms in equal measure. Furthermore, they should integrate an understanding of online antisemitism as part of a wider ecosystem of highly interconnected networks. 

Conclusions  

The Bondi Beach attack was the product of an online antisemitism environment that is more organisationally diffuse and more transnationally interconnected, and that prompts a wider range of harms than regulatory systems are currently equipped to address. The online threat picture comprises a nebulous web of hate, extremist and conspiracy movements, often boosted by hostile states. Diverse ideological movements – which previously had little in common – are increasingly united by a shared antisemitic core.   

The sheer volume, coded nature and rapid pace of change of antisemitic content have left platform responses struggling to keep up. Meanwhile, platform systems, such as recommender algorithms, serve to exacerbate the issue by surfacing antisemitic content to users. New technologies, such as generative AI, have been leveraged by hostile states to further manipulate the online information environment. These trends are global in origin but have direct, measurable consequences not only for community safety but also social cohesion and democratic integrity in Australia. 

The hybrid nature of the contemporary online antisemitism threat requires policy responses that integrate, rather than silo, the range of actors, ideologies and harms involved. Regulatory frameworks designed around group-based proscription, platform-by-platform content removal or discrete harm categories will continue to miss most of the antisemitic content that drives real-world harm. Effective responses must address online antisemitism as a wider ecosystem of highly interconnected networks. They must respond to the cyclical relationship between online incitement and offline violence. Rather than simply responding to the sharp tip of antisemitism, responses must understand and address the long tail of harms associated with online antisemitism. 

Transparency is the foundational precondition for every other accountability mechanism. Platforms cannot be held accountable for systems that cannot be independently examined; regulators cannot assess the effectiveness of mitigation measures that are not disclosed; communities cannot identify failures in protections they cannot see. The EU Digital Services Act (DSA) has demonstrated that mandated, legally enforceable transparency obligations are achievable, and that they also produce unprecedented public insight into how platforms handle antisemitic content. The remaining gaps (such as cross-platform comparability, antisemitism-specific enforcement data and independent researcher data access) define the agenda for the next generation of regulatory development. 

Several design principles consistently distinguish more from less effective regulatory approaches. Systemic obligations outperform reactive removal mandates: the German Network Enforcement Act (NetzDG) demonstrates that complaint-triggered takedown laws cannot reach the platform architecture driving harmful content. Voluntary approaches are insufficient as a primary mechanism, as demonstrated by the Christchurch Call’s experience and the history of the EU hate speech code of conduct. And scope matters as much as strength: even rigorous enforcement against clearly illegal content leaves most harmful material (such as coded antisemitism or borderline extremist content) unaddressed unless systemic risk frameworks reach further. The EU DSA and UK Online Safety Act, taken together, currently represent the most advanced regulatory frameworks for addressing these harms: they have created the accountability infrastructure that makes measuring and improving effectiveness possible over time. Australia has an opportunity, through this Royal Commission and the legislative and regulatory work that follows, to build on the lessons of these frameworks and establish a regulatory approach commensurate with the scale, sophistication and urgency of the online antisemitism threat. 

To read ISD’s full submission to the Royal Commission, including further analysis of the online antisemitism landscape and global regulatory approaches, click here.   

Annex 1: LLM Classification Methodology 

To identify antisemitic social media activity, ISD employed a bespoke LLM trained by subject matter experts with a deep understanding of the nuanced ways that antisemitic rhetoric presents in online discourse. The LLM was trained on a high threshold application of the IHRA definition of antisemitism, taking into account context and ISD’s wider definition of hate. Anthropic’s Claude Haiku was selected based on high performance, with an F1 score of 0.87. An F1 score measures a model’s performance and is calculated from two metrics: precision (the proportion of posts labelled as antisemitic by the model that were, in fact, antisemitic) and recall (the proportion of antisemitic posts that the model correctly identified). Classifier performance was validated by manually coding a randomly sampled subset of posts and comparing results against model outputs.