Israel using AI to pinpoint Hamas leaders, find hostages in Gaza tunnels — report
AI programs, including chatbot and audio-based location tool, reportedly developed by Unit 8200 soldiers and reservists from big tech firms, said to have raised ethical concerns
Israel has made unprecedented use of artificial intelligence to wage war in Gaza, with many programs developed through cooperation between enlisted intelligence soldiers and reservists who work at large tech firms, The New York Times reported Friday, citing European, American and Israeli defense officials.
While Israel’s use of AI in warfare has previously been revealed, the latest report details several different ways the IDF has incorporated AI, including a chatbot trained in multiple Arabic dialects that has served the IDF to gauge public sentiment, an AI-infused audio program that pinpoints targets based on sounds “such as sonic bombs and airstrikes,” and a facial recognition program that matches hidden or injured faces to real identities.
Israeli officers cited by the Times said the AI helped find hostages and sped up certain tasks, but acknowledged the applications could sometimes be faulty. According to the Times, “some officials have struggled with the ethical implications of the AI tools” because they can lead to increased surveillance, civilian deaths and wrongful arrests.
Responding to the Times, the IDF said specific technologies were confidential and could not be commented on. The newspaper quoted the military as saying Israel “is committed to the lawful and responsible use of data technology tools.”
Four Israeli officials cited by the Times said the IDF quickly cleared AI tools for deployment after the Gaza war was sparked on October 7, 2023, when thousands of Hamas-led terrorists stormed southern Israel to kill some 1,200 people and take 251 hostages.
Some uses of AI in Gaza have been previously reported, including a virtual reality system that helps soldiers scan urban war zones. Israeli-Palestinian activist outlet +972, which first reported on the Arabic-language chatbot, has also reported on a program called Lavender, said to have helped the IDF amass a list of 37,000 human targets based on their ties to Hamas. The military has denied using AI to generate such a kill list.
Many of the AI tools were developed in an innovation hub set up by the Military Intelligence Directorate’s Unit 8200, according to the Times. Known as “The Studio,” the hub helped connect enlisted soldiers with reservists who work at companies such as Meta, Google and Microsoft, the Times said, citing the four Israeli officials.
Avi Hasson, who heads the nonprofit Startup Nation Central that connects investors with companies in Israel, told the Times that the reservists “brought know-how and access to key technologies that weren’t available in the military.”
Meta and Microsoft declined to comment, while Google responded that it had reservist employees in numerous countries and that their work was unrelated to the company, the Times said.
Citing three Israeli and American officials briefed on the matter, the newspaper said the IDF in October 2023 infused AI into decade-old technology in a bid to pinpoint Ibrahim Biari, the commander of Hamas’s Central Jabalia Battalion.
The new tool reportedly enabled the IDF to track Biari’s phone calls to a Hamas tunnel complex under northern Gaza’s densely populated Jabalia refugee camp. A strike on the area was greenlit despite warnings within the military that several apartment buildings would need to be targeted to ensure Biari was killed. The strike, said to have killed dozens of people, is being probed by the military, which told the Times it was “unable to provide any further information until the investigation is complete.”
The audio tool used to locate Biari has been sharpened over time to more accurately locate people, and has served the IDF in finding hostages in Gaza, the Times said, citing Israeli officers.
Following the Hamas onslaught, Israel also erected temporary checkpoints between the Strip’s north and south, with security cameras capable of sending high-resolution images to an AI facial recognition program, the Times said.
Two Israeli intelligence officers cited by the newspaper said the system sometimes had trouble identifying obscured faces, leading to the arrest and interrogation of Palestinians flagged by mistake.
Three Israeli officers cited by the Times said “The Studio” at Unit 8200 also developed an AI large language model capable of handling texts in various dialects of spoken Arabic.
According to the Times, such tools are difficult to develop because most relevant training data appears in standard written Arabic. The IDF, on the other hand, “has decades of intercepted text messages, transcribed phone calls and posts scraped from social media in spoken Arabic dialects,” the Times said, citing the three officers.
The program was said to have been developed in the early months of the war and incorporated into multimedia databases, enabling military analysts to run complicated searches of images and videos, the Times said.
The chatbot reportedly had difficulty identifying modern slang and words transliterated from English, requiring intelligence officers with knowledge of the dialects to check its output. The program was also said to have provided some wrong answers, such as “returning photos of pipes instead of guns.”
Nonetheless, according to two Israeli intelligence officers cited by the Times, the chatbot has significantly sped up the military’s research and analysis.
The Times cited three officers as saying the technology allowed the military to gauge whether there was public pressure in the Arab world for a counterstrike on Israel after it assassinated Hezbollah chief Hassan Nasrallah in September.
The Times of Israel Community.