Israeli media has reported that Jonathan Orich, an aide to Prime Minister Benjamin Netanyahu, developed a system using artificial intelligence to monitor media reports related to Netanyahu and his wife Sara. The system, which was designed to scan around 50 news sources every 90 seconds, was created using the Claude AI engine from Anthropic. It sent summaries of reports to two WhatsApp groups, one focused on Netanyahu and his party, and the other on Sara Netanyahu.
The tool was operational from at least September 1 and conducted over 19,000 scans of media reports by Thursday. It monitored prominent Israeli news sites, as well as accounts of senior journalists and dozens of Telegram channels and accounts publishing open-source intelligence. The system was provided with a list of keywords to track, including names of Netanyahu, his wife, and other top Israeli politicians, as well as around 12 Israeli political parties.
The AI tool assessed reports based on a specific index created by Orich to determine their importance, prioritizing coverage from major media outlets such as Channel 12, Yedioth Ahronoth, and the Israeli Broadcasting Corporation. It also monitored and summarized election polls published by Israeli media. However, a security flaw was discovered when Orich uploaded the tool's code to GitHub, leaving it unencrypted and publicly accessible.
The exposed code revealed sensitive information, including private phone numbers of senior officials. The tool also used an unencrypted access code, which could have allowed anyone to access WhatsApp group data and send messages through the tool. After being contacted by Haaretz, Orich restricted his GitHub account and removed the tool from public view.
The AI system was not limited to monitoring and summarizing news; it was also programmed to prepare proposed media responses to reports that required comment. The code showed that the tool was supposed to identify news stories that needed a response, formulate a precise and realistic message, and determine whether a response was necessary and how to respond.
However, a key component of the tool was found to be faulty, with data showing it experienced several malfunctions and failed to execute its task more often than not. This development comes as Orich faces other legal issues, including charges related to transferring sensitive information with the intention of harming Israel's security.
Orich is facing a case from June 11, where he was charged with taking sensitive information with the aim of harming Israel's security. This relates to his role in leaking a confidential military intelligence document to the German newspaper Bild in September 2024. The use of AI in monitoring media reports raises questions about the extent of such practices and the potential risks associated with them.
Key points
- An aide to Israeli Prime Minister Benjamin Netanyahu developed an AI tool to monitor media reports on Netanyahu and his wife.
- A security flaw in the tool exposed sensitive information, including private phone numbers of senior officials.
- The tool was designed to prepare proposed media responses to reports that required comment, but a key component was found to be faulty.
