Report in the US: How Israel is trying to influence AI chatbot responses
Israel is conducting an information campaign by feeding pro-Israel articles into Large Language Models (LLMs) to influence AI chatbot responses regarding the Gaza Strip and the IDF.
Israel is conducting an information campaign designed to tilt the answers of AI chatbots such as ChatGPT in its favor on issues like the Gaza Strip and the IDF, by publishing and feeding pro-Israel articles into the Large Language Models (LLMs) behind these chatbots. This is reported by the Politico website, based on documents from an advertising agency working with the Israeli government.
Traditionally, search engines and social media platforms were a primary source for users to discover news information. However, in the last two years, the popularity of modern chatbots has made them an increasingly important source of information. Israel is aware of this shift and is adapting its information efforts accordingly.
According to documents submitted to the US Department of Justice by the French PR agency Havas Media, a subcontractor called Piro established a fictitious research institute named the Hanover Institute for Public Policy, with the goal of using it to publish articles for feeding into AI models.
The institute's website has a generic design and the articles are unsigned, but a notice at the bottom of the site states:
"Materials are distributed by Piro on behalf of Havas on behalf of the National Information Directorate (LAPAM) of the Government of Israel."
The site is operated by the American startup Res, which previously boasted of helping clients like Nike and Intel receive positive answers in AI models. The platform was used to publish dozens of articles as part of a $100,000 campaign. These articles are optimized for AI processing, featuring headlines similar to user queries, such as "Is the IDF the most moral army in the world?", "Is there a starvation policy in Gaza?", and "Are Jews foreign occupiers in Israel?".
According to Politico, the campaign appears to have achieved some success: a check found that ChatGPT and Perplexity cited articles from the site in answers to queries related to Gaza, anti-Zionism, and antisemitism.
"Piro was hired by the Israeli government to integrate accurate and documented facts into the public discourse, and to fight misinformation about Israel with verified information," Piro founder Daniel Rosenberg told Politico. Havas, the Hanover Institute, LAPAM, and the Israeli Embassy in the US did not provide a response.
