By: Fern Sidman- Jewish Voice News

Large language models marketed as “artificial intelligence” were once heralded as tools that could democratize knowledge, accelerate innovation, and connect societies across borders. Yet new research indicates that the same technologies are now being aggressively exploited by extremists who openly advocate violence against Jews, exposing a dangerous vulnerability in the global AI ecosystem. According to information provided by the Anti-Defamation League (ADL) and cited by The Algemeiner in a report on Thursday, the proliferation of open-source AI models has created fertile ground for antisemitic propaganda, genocidal rhetoric, and coordinated digital hate campaigns—often with minimal technical barriers and few effective safeguards.

The findings arrive amid growing concern among security experts, civil-rights organizations, and governments that artificial intelligence is rapidly becoming an accelerant for extremism rather than a neutral tool. As The Algemeiner reported, watchdog groups now warn that the speed at which AI capabilities are advancing has outpaced the development of ethical controls, leaving Jewish communities—and democratic societies more broadly—exposed to unprecedented forms of digital incitement.

On Tuesday, the ADL released a comprehensive study titled “The Safety Divide: Open-Source AI Models Fall Short on Guardrails for Antisemitic, Dangerous Content.” The report analyzed 17 widely available large language models (LLMs), including Google’s Gemma-3, Microsoft’s Phi-4, and Meta’s Llama 3—systems that can be freely downloaded, modified, and customized by users.

As The Algemeiner report summarized, the results were deeply troubling. Researchers found that many of these open-source models could be easily manipulated to generate antisemitic narratives, conspiracy theories, and content advocating mass violence against Jews. In some cases, minimal prompting was required; in others, the models could be “fine-tuned” by users to produce extreme content with chilling consistency.

“The ability to easily manipulate open-source AI models to generate antisemitic content exposes a critical vulnerability in the AI ecosystem,” ADL CEO Jonathan Greenblatt said in a statement highlighted in The Algemeiner report. “The lack of robust safety guardrails makes AI models susceptible to exploitation by bad actors, and we need industry leaders and policymakers to work together to ensure these tools cannot be misused to spread antisemitism and hate.”

The ADL’s findings underscore a central dilemma in modern AI development: while open-source models promote transparency and innovation, they also decentralize responsibility, making it difficult to enforce ethical standards or prevent malicious use.

To better understand the scale of the problem, ADL researchers compared open-source models with so-called “closed-source” systems operated by major companies, including OpenAI’s GPT-4o and GPT-5. As The Algemeiner reported, the results defied expectations.

OpenAI’s GPT-4o outperformed nearly every open-source model in terms of safety, refusing to generate harmful content in most test cases and consistently flagging antisemitic prompts. By contrast, GPT-5—despite being a newer model—performed noticeably worse across several benchmarks.

According to the ADL’s analysis, GPT-5 had a lower overall guardrail score (.75 compared to .94 for GPT-4o), fewer outright refusals (69% versus 82%), and, most alarmingly, produced harmful content in 26% of cases where GPT-4o generated none. The newer model also exhibited a higher evasion rate, meaning it was more likely to be coaxed into bypassing safety restrictions.

The findings, cited in The Algemeiner report, raised difficult questions about how AI safety is measured and implemented. Researchers suggested that GPT-5’s design philosophy—favoring “safe completions” rather than direct refusals—may have contributed to the results. Instead of declining to answer sensitive prompts, GPT-5 often addressed them at a high level, sometimes illustrating antisemitic tropes in the process.

In one example described in the report, GPT-4o prefaced its response with a warning about the sensitivity of the topic, while GPT-5 omitted such disclaimers and instead embedded problematic stereotypes within its explanation.

Despite these concerns, the ADL cautioned against drawing overly simplistic conclusions. “We cannot claim a strict linear boost in overall capability,” the researchers noted, acknowledging the complexity and ambiguity inherent in evaluating AI behavior across varied contexts.

While the ADL focused on the vulnerabilities of mainstream AI models, a parallel investigation by the Middle East Media Research Institute (MEMRI) revealed how extremists are already capitalizing on those weaknesses. As The Algemeiner reported, MEMRI recently uncovered a growing ecosystem of custom-built AI tools explicitly designed to promote neo-Nazi ideology.

Among the most disturbing discoveries were chatbots branded as “Fuhrer AI” and “Deep AI Adolf Hitler Chat,” programmed to emulate the speech patterns of the Nazi leader and glorify his genocidal worldview. These systems, MEMRI found, are not isolated curiosities but part of a broader trend in which extremists are building tailored AI personas to recruit followers, spread propaganda, and normalize calls for violence.

“We are also witnessing the rise of a new digital infrastructure for hate. And it’s not just fringe actors,” wrote Steven Stalinsky, MEMRI’s executive director, and Simon Purdue, director of MEMRI’s Violent Extremism Threat Monitor, in an analysis cited by in The Algemeiner report. “State-aligned networks from Russia, China, Iran, and North Korea amplify this content using bots and fake accounts, sewing division, disinformation, and fear—all powered by AI. This is psychological warfare. And we are unprepared.”

Their warning reflects a growing consensus among analysts that AI-driven hate campaigns are no longer spontaneous or amateurish. Instead, they are increasingly coordinated, well-funded, and integrated into broader information-warfare strategies pursued by hostile state actors.

The danger posed by AI-enabled extremism did not emerge overnight. MEMRI researchers noted that jihadist and neo-Nazi groups began experimenting with generative AI as early as 2022. Since then, the volume, sophistication, and coordination of AI-assisted propaganda have expanded dramatically.

As The Algemeiner report documented, extremist groups now use AI to generate multilingual content at scale, tailor messages to specific audiences, fabricate images and videos, and automate online harassment campaigns. These capabilities allow them to overwhelm moderation systems and create the illusion of widespread support for hateful ideologies.

The ADL and MEMRI’s findings complement one another, painting a comprehensive picture of how AI is reshaping the threat landscape. In October, the ADL released a separate report titled “Innovative AI Video Generators Produce Antisemitic, Hateful, and Violent Outputs,” while MEMRI last month published “Artificial Intelligence and the New Era of Terrorism,” an in-depth assessment of how jihadist groups are leveraging AI to enhance recruitment and operational planning.

As The Algemeiner noted in its coverage, these reports collectively suggest that the challenge is not merely technical but civilizational, raising urgent questions about governance, accountability, and resilience in the digital age.

In response to its findings, the ADL issued a series of policy recommendations aimed at curbing the misuse of AI. Governments, the organization argued, should establish strict controls on the deployment of open-source AI in official settings, mandate independent safety audits, and require collaboration with civil-society experts specializing in hate and extremism.

The ADL also called for clear disclaimers on AI-generated content addressing sensitive topics, a measure designed to prevent users from mistaking algorithmic outputs for authoritative or neutral information. As The Algemeiner report emphasized, these proposals reflect growing concern that AI systems are being treated as neutral tools when, in practice, they shape narratives and perceptions at scale.

Daniel Kelley, director of the ADL’s Center for Technology and Society, framed the issue in stark terms. “The decentralized nature of open-source AI presents both opportunities and risks,” he said. “While these models increasingly drive innovation and provide cost-effective solutions, we must ensure they cannot be weaponized to spread antisemitism, hate, and misinformation that puts Jewish communities and others at risk.”

While civil-society groups call for stronger safeguards, Israel has begun taking concrete steps to address the security implications of AI. As The Algemeiner reported, the Israel Defense Forces recently announced the launch of its “Bina” initiative—named after the Hebrew word for “intelligence”—a sweeping effort to integrate artificial intelligence into military planning and operations.

The initiative aims to consolidate Israel’s AI-driven capabilities across intelligence, cyber defense, and battlefield decision-making, with a particular focus on countering threats from Iran, China, and Russia. Israeli officials have signaled that AI will play a central role not only in conventional warfare but also in combating information operations and psychological warfare conducted online.

The move reflects a broader recognition within Israel’s security establishment that AI has become a strategic domain in its own right. As The Algemeiner report noted, the same technologies that enable rapid data analysis and operational efficiency can also be exploited by adversaries to spread disinformation, incite violence, and undermine social cohesion.

The convergence of findings from the ADL, MEMRI, and Israeli security planners points to a sobering conclusion: artificial intelligence is amplifying both human creativity and human malice. Without robust guardrails, the technology risks becoming a force multiplier for some of the oldest and most dangerous hatreds.

For Jewish communities worldwide, the implications are immediate and personal. AI-generated antisemitic content can circulate at unprecedented speed, reaching audiences that traditional Jew hating propaganda never could. For democratic societies, the stakes are equally high, as AI-driven disinformation threatens to erode trust, polarize populations, and normalize calls for violence.

As The Algemeiner has consistently argued in its coverage of antisemitism and emerging technologies, the challenge now facing policymakers, tech companies, and civil-society organizations is not whether AI will shape the future—it already has—but whether that future will be governed by ethical restraint or surrendered to those who seek to weaponize innovation for destruction.

In the words of MEMRI’s analysts, the threat is neither theoretical nor distant. It is already here, embedded in code, circulating through networks, and reshaping the digital public square. The question, as The Algemeiner report framed it, is whether the world will respond with urgency and coordination—or continue to underestimate the cost of inaction until the damage is irreversible.