NEW YORK – The Anti-Defamation League said in recent posts that text-to-video artificial intelligence generators produced antisemitic, extremist, or hateful outputs when prompted in roughly 40 percent of its tests. The ADL called the results alarming and urged immediate action from developers and platforms.
The finding comes amid broader evidence that extremists and malign actors are increasingly using AI tools to automate and amplify hate and disinformation. A separate analysis by security researchers and reporting by mainstream outlets documented how AI can create convincing deepfakes and manipulate visuals and audio to spread extremist narratives.
ADL researchers said they used a range of prompts designed to test for antisemitic tropes, glorification of violence, Holocaust denial, and other hateful content. In many cases, the text-to-video systems returned content that echoed classic antisemitic themes or depicted violent imagery tied to Jewish targets. The organization stressed that these were not synthetic edge cases but repeated failures across multiple prompts.
Advocates and security experts warned that generative video is uniquely dangerous because it combines visual realism with audio and narrative. Videos can be reposted and monetized on social platforms and messaging channels. That speed and reach make harmful content easier to weaponize against Jewish communities and pro-Israel voices, they said.
The problem is not limited to video. The ADL has previously documented bias and troubling responses in large language models and other generative systems when asked about Jews and Israel. Those earlier findings prompted calls for transparency in training data and for third-party audits of models. The new text-to-video results add urgency to those demands.
Industry responses have been mixed. Some AI developers point to safety filters and content moderation tools as part of their defense. Critics say current safeguards are inconsistent and are failing to catch obvious references to antisemitic tropes such as blood libel or Holocaust denial. Advocates urged companies to pause open deployments of text-to-video tools until robust, independently audited protections are in place.
Jewish and pro-Israel organizations called the ADL findings a wake-up call. They said platforms and regulators must treat generative video as a major vector for modern antisemitism. Groups urged clearer reporting pathways, faster takedown of violent or dehumanizing content, and funding for research to develop better detection tools targeted at visual antisemitic narratives.
Legal experts said the research could influence policymakers weighing new rules for AI. Potential measures include mandatory risk assessments before deployment, transparency requirements about datasets and training methods, and stricter liability for platforms that host or amplify illegal hate content. Some called for specific protections for targeted groups, including Jewish communities, in any new regulatory frameworks.
For Jewish and pro-Israel readers, the implications are clear. Generative video that normalizes or amplifies antisemitic messages can seed offline violence, intimidate Jewish students and communities, and erode trust in public discourse. Community leaders urged vigilance and collaboration with law enforcement and tech firms to ensure hateful content is identified and removed promptly.
The ADL recommended immediate steps for companies and policymakers. These include publishing red-team test results, strengthening multimodal content filters, and working with civil society to build targeted defenses against hate that appears in images, audio, and video. The organization also urged public education campaigns so users can better spot and report AI-generated hate.
This article is based on verified information from the Anti-Defamation League, Secure Communities Network, CBS News, PBS NewsHour, and reporting on AI and hate from Homeland Security Newswire.