Israel is developing a ChatGPT-style artificial intelligence tool that weaponizes mass surveillance against Palestinians, marking a dangerous escalation in the use of AI for military and intelligence operations. According to an in-depth investigation by +972 Magazine, Local Call, and The Guardian, the Israeli army is training an advanced AI model using vast amounts of intercepted conversations in Arabic, gathered through sweeping surveillance mechanisms in the occupied Palestinian territories.
This new AI system, developed under the direction of Unit 8200—Israel’s elite cyber intelligence division—functions as a Large Language Model (LLM), a sophisticated machine-learning tool capable of analyzing, translating, and generating text at scale. Unlike commercial models such as ChatGPT, which are trained on publicly available internet data, this Israeli military model is being fed an immense database of intelligence obtained through the systematic monitoring of Palestinian civilians. The project underscores how AI is being integrated into an already pervasive surveillance apparatus, further entrenching Israeli control over Palestinian lives.
Three Israeli security officials, speaking to +972 Magazine, Local Call, and The Guardian, confirmed the existence of the AI tool. As of late 2023, the model was still being trained, and while its full operational scope remains unclear, military sources suggest its primary function will be to rapidly analyze intercepted material, enabling Israeli forces to profile individuals, generate intelligence reports, and accelerate the process of incrimination and arrest. Given Unit 8200’s existing use of smaller language models for identifying so-called “troublemakers” based on their speech, this new tool could exponentially increase the volume of Palestinians targeted for detention.
“AI magnifies power,” an intelligence source told +972. “It allows the system to process and act on the data of far more people than before. With a tool like this, I can track human rights activists, I can monitor Palestinian construction in Area C, I can see who is speaking to whom. When you possess this level of intelligence, you can manipulate it for any purpose you choose.”
Though the AI program’s development predates the current war in Gaza, its evolution accelerated following the October 7 attacks. In response to the war, Unit 8200 enlisted Israeli citizens with expertise in machine learning from major tech companies like Google, Meta, and Microsoft, seeking their knowledge to refine its language models. One Israeli security source described the influx of tech professionals as bringing capabilities “previously accessible only to a very exclusive group of companies worldwide.” While Google acknowledged that some of its employees serve in military reserve units, the company insisted that such work is separate from its operations. Meta and Microsoft declined to comment.
The scale of the surveillance underpinning this AI model is staggering. Sources revealed that Unit 8200’s chatbot has been trained on 100 billion words of spoken Arabic, much of it gathered from intercepted phone calls, text messages, and digital communications. Human rights experts warn that this constitutes an extreme violation of privacy and digital rights. Zach Campbell, a senior researcher at Human Rights Watch, described it as “highly personal information, taken from people who are not suspected of any crime, to train a tool that could later be used to justify their arrest.”
Nadim Nashif, founder of the Palestinian digital rights group 7amleh, underscored the broader implications: “Palestinians are being used as test subjects in Israel’s AI laboratory. These technologies, developed under the pretense of security, are ultimately tools of apartheid, designed to maintain an oppressive system of control. This is a flagrant violation of Palestinian human rights.”
The existence of the AI tool was first hinted at by Chaked Roger Joseph Sayedoff, an Israeli intelligence officer who presented himself as the project’s lead. During a little-noticed 2023 lecture at the DefenseML conference in Tel Aviv, he detailed how the military sought to build “the largest dataset possible” of Arabic-language intelligence. He also admitted that standard language models could not process Arabic dialects, requiring Israel to develop its own LLM “based on the dialects that hate us.”
Unit 8200’s drive to create this tool was initially hindered when OpenAI rejected the Israeli military’s request for direct access to its LLM, preventing its integration into Israel’s classified intelligence systems. However, the military later obtained OpenAI’s language model through Microsoft Azure, as revealed in a separate investigation by +972 Magazine and Local Call.
AI experts warn that such technology, particularly when deployed for military purposes, is highly prone to error. Large Language Models are probabilistic systems, meaning they generate responses based on statistical likelihoods rather than absolute truths. This can result in what AI researchers call “hallucinations”—false or misleading conclusions presented as factual insights. Despite these risks, intelligence agencies tend to place an excessive degree of trust in AI-generated outputs, increasing the likelihood of wrongful arrests and unjustified targeting.
Previous investigations by +972 Magazine have already exposed the Israeli army’s use of AI-driven targeting programs such as Lavender, which autonomously generated “kill lists” during Israel’s bombing campaign in Gaza. The system flagged tens of thousands of Palestinians as potential militants based on behavioral patterns, despite an acknowledged 10% error rate. Human oversight, according to sources, was little more than a “rubber stamp,” and soldiers carried out Lavender’s recommendations as if they were definitive human decisions.
This latest AI tool represents an even greater expansion of Israel’s digital occupation. By enabling mass profiling and preemptive arrests based on intercepted speech, the Israeli military is shifting toward a system of predictive policing in which dissent can be criminalized before it even manifests in action. The consequences for Palestinians are dire: With an ever-expanding database of monitored conversations, individuals can be detained indefinitely under administrative detention—Israel’s system of holding prisoners without charge or trial.
A former intelligence officer bluntly described the army’s approach: “Sometimes, it’s just a division commander who wants 100 arrests per month in his area.”
Unit 8200’s strategy is clear: total surveillance, total control. By feeding its AI with intercepted Arabic conversations—many of which have no direct intelligence value—the military is creating a system that can be directed against any segment of Palestinian society, from activists to ordinary civilians. The result is a future in which an algorithm, not a human, decides who is a threat, who is a suspect, and who is next in line for arrest.
Israel’s reliance on AI-driven surveillance may be unprecedented in scale, but it is not an isolated phenomenon. Intelligence agencies in the U.S. and U.K. are also developing LLM-based tools for analyzing vast amounts of data. However, former Western intelligence officials told +972 Magazine, Local Call, and The Guardian that Israel’s approach is significantly more aggressive, with fewer ethical safeguards and far greater risks of abuse.
Brianna Rosen, a former White House security official and researcher at Oxford University, explained that while AI-enhanced intelligence analysis can uncover patterns that human analysts might miss, it also introduces a high potential for error. “Mistakes will be made, and those mistakes could have devastating consequences,” she said.
Israeli military sources, however, appear unconcerned with accuracy. The real objective, they suggest, is volume. With an endless supply of intercepted conversations and a rapidly evolving AI system, the goal is not just to monitor Palestinians—it is to control them at an unprecedented scale.
The Israeli army declined to answer specific questions about the AI project, citing its sensitive nature, but insisted that “any use of technological tools is done through a rigorous process led by professionals, in order to ensure maximum accuracy of the intelligence information.”
The reality, as uncovered by +972 Magazine and its partners, suggests otherwise. As AI-driven surveillance becomes an integral part of Israel’s military strategy, the line between intelligence gathering and digital colonialism is becoming increasingly indistinguishable. For Palestinians, the consequences are already being felt—and the future promises even greater oppression, powered by algorithms designed to watch, analyze, and control every aspect of their lives.
