Israel’s Covert Campaign to Shape ChatGPT’s Gaza Answers Exposes New AI Influence Front
Documents filed with the U.S. Justice Department indicate Israel has been funding a covert effort to influence how language models like ChatGPT respond to questions about Gaza and the Israeli military. The campaign turns AI systems into a new information battleground, raising fresh questions for tech firms, regulators and users about who is quietly trying to script what the world’s chatbots say about war and human rights.
Israel is bankrolling a covert influence campaign aimed at shaping how large language models such as ChatGPT answer questions about the war in Gaza and the conduct of Israeli forces, according to documents submitted to U.S. authorities. The effort pulls artificial intelligence tools into the heart of one of the world’s most contested narratives, turning chatbots into targets of the same information wars long waged on social media and traditional news.
The filings, summarized by Spanish‑language reporting on 15 August, describe an Israeli‑funded operation designed to steer responses from widely used AI systems in ways more favorable to Israel at a time when global public support for its actions has been slipping. The documents were lodged with the U.S. Department of Justice under foreign influence disclosure rules, indicating that those running the campaign acknowledged they were acting on behalf of a foreign principal.
While the filings’ full technical details are not public, the description points to a mix of tactics: coordinated user interactions meant to “train” or nudge AI models through feedback mechanisms, the seeding of preferred narratives across the internet sources these systems ingest, and potentially the crafting of prompts and content designed to exploit known weaknesses in how large language models weigh and reproduce information. The stated goal is to influence how these models frame questions about Gaza, Israeli military operations, and allegations of human rights abuses.
For ordinary users, the stakes are largely invisible but profound. Many now turn to AI chatbots for first‑pass explanations of conflicts and crises, often without checking primary sources. If a foreign government can systematically tilt those answers—by amplifying certain facts, downplaying others, or reframing legal and moral debates—it effectively gains a new channel to reach global audiences, one that feels neutral and authoritative because it comes from a “machine.” Civilians in Gaza, Israelis, diasporas, and activists worldwide all risk seeing their experiences filtered through narratives influenced not just by open political campaigns, but by quiet interventions in the tools they rely on for understanding.
The campaign also poses hard questions for technology companies building and operating these models. They already struggle to manage bias, misinformation, and state‑backed manipulation on their social platforms. Now they face a new front: sophisticated attempts to game the training and reinforcement processes of AI systems that produce fluent, confident‑sounding answers even when the underlying evidence is contested. Existing safeguards—such as content moderation policies, red‑teaming, and data curation—were not designed with coordinated foreign narrative‑shaping in mind.
Strategically, Israel’s reported effort is an early example of what many security experts have warned about: that states will treat large language models as high‑value terrain in psychological and information operations. If one country is documented as trying to steer ChatGPT‑style systems on a specific conflict, others are likely to follow suit on issues from territorial disputes to election interference and sanctions narratives. AI companies, regulators, and democratic institutions are only beginning to define what kinds of influence operations on these systems are unacceptable, how to detect them, and how to respond when they originate from close allies rather than adversaries.
The core insight is that influence campaigns no longer need to change what people see in their news feeds; they can now aim to change what people think “the neutral AI” believes.
In the near term, watch for whether U.S. regulators or lawmakers demand more transparency from AI providers about foreign attempts to shape model outputs, and whether tech firms publish independent audits of how their systems answer questions on Gaza and similar flashpoints. Other key signals will be whether additional foreign‑funded AI influence operations surface in Justice Department filings, and if multilateral forums begin treating the integrity of large language models as part of the broader debate over disinformation, digital sovereignty, and the laws of war in the information age.
Sources
- OSINT