Social media influencer Samantha Ettus has noticed a troubling change since the October 7, 2023 Hamas attack on Israel. Her platforms, which advocate for Israel and Jewish communities, are now flooded with antisemitic bot messages targeting her followers. Blocking these accounts has become a daily battle that takes up hours of her time. “They come in fast and furious,” Ettus explains. “The amount of time I spend blocking accounts is truly outrageous. I have to do it because otherwise I personally feel an obligation to people who follow me.”
Her experience reflects a growing problem that researchers say is being supercharged by artificial intelligence. Large language models that power AI chatbots are being manipulated to create and spread hateful content faster than ever before.
The AI Amplification Problem
Ashique KhudaBukhsh, a computer scientist at Rochester Institute of Technology who studies these systems, puts it simply: “AI has made it possible to scale up any kind of inaccurate information you can and generate it very fast.” The technology allows antisemitic content to be created and distributed across the internet without human involvement, making the problem exponentially worse.
The issue isn’t limited to fringe platforms. Even major AI systems have shown concerning biases. Elon Musk’s Grok AI chatbot on X was caught delivering antisemitic responses to users in July 2023, just weeks after Musk said he wanted it “retrained” because he thought it was too politically correct. The company later said it took action to ban hate speech and improve the model’s training.
Major AI Companies Under Scrutiny
Research from the Anti-Defamation League’s Center for Technology and Society examined four leading AI systems—ChatGPT, Claude, Google’s Gemini, and Meta’s Llama, and found all of them reflected bias against Jews and Israel. The study particularly criticized Llama, Meta’s open-source model, for scoring lowest on both bias and reliability measures.
Meta pushed back on the findings, arguing that the research methodology didn’t reflect how people actually use AI tools. “People typically use AI tools to ask open-ended questions that allow for nuanced responses, not prompts that require choosing from a list of pre-selected multiple-choice answers,” a company spokesperson said.
How Deep Does the Problem Go?
KhudaBukhsh’s research team discovered that AI models can be easily manipulated to produce antisemitic content simply by asking them to make previous statements “more toxic.” The results included calls for ethnic cleansing, Holocaust denial, and other deeply harmful content.
The root of the problem lies in the training data these AI systems learn from. “The models are learning all these things from the data,” KhudaBukhsh explains. When training data contains problematic content that describes certain groups as subhuman, the AI can form dangerous connections about eliminating those groups.
This bias extends beyond obvious hate speech into subtler discrimination. AI systems used in hiring might unfairly reject candidates with Jewish-sounding names, showing how these biases can affect real-world decisions.
Advocates argue that current laws aren’t equipped to handle AI-generated content. Section 230 of the 1996 Communications Decency Act protects tech companies from liability for user-generated content, but it’s unclear how this applies to AI-created material.
Yaël Eisenstat from Cybersecurity for Democracy believes companies won’t self-regulate effectively. “They are not incentivized legally, they are not incentivized by their investors, they are not incentivized politically,” she says.
The stakes are rising rapidly. The global market for large language models is expected to jump more than 530% by 2030, reaching $35.4 billion according to Grand View Research.
A Generational Shift
Perhaps most concerning is how younger people increasingly use chatbots instead of traditional search engines. “People are pulling up ChatGPT the way they are using Google,” notes Daniel Kelley from the ADL. This shift means AI biases could fundamentally shape how an entire generation views the world.
Though AI-generated images and film are becoming more realistic, scientists raise warning flags about a “ticking clock” to control these technologies prior to even more powerful models entering markets without protection.
The challenge is plain: as AI continues to grow in power and reach, the need for regulation and correction of bias is more significant than ever.
