How Iran’s military harnesses ChatGPT

{"text":[[{"start":6.5,"text":"As the Middle East waits on fraught peace negotiations, Iran’s embattled hackers have been hunched over their laptops, chatting to a game-changing new weapon: US-made AI models. "}],[{"start":18.3,"text":"ChatGPT, Gemini and other western AI programmes have turbocharged Iran’s cyber operations, helping them develop malware, craft phishing messages in perfect Hebrew and Arabic, and then unleash attacks at unprecedented scale and speed, cyber experts and tech companies say.  "}],[{"start":36.6,"text":"“We are seeing signs that they are using AI prompts the entire way,” said a cyber security analyst at a big firm, speaking anonymously to discuss the sensitive issue. “It has absolutely helped them raise their game.” "}],[{"start":48.95,"text":"This has enabled Iran, which has been in a fragile ceasefire with Israel and the US since early April, to keep up the digital pressure on its more advanced adversaries, scanning the internet for enemy vulnerabilities and safeguarding Tehran’s own weaknesses. They even use AI to create convincing personas to dupe targets in the US and Israel."}],[{"start":69.7,"text":"The United Arab Emirates, battered by thousands of missiles and drones during the fighting, has said it was facing more than half a million cyber attacks every day, assisted by OpenAI’s ChatGPT. Israelis have been spammed with wave after wave of phishing emails and texts, some reportedly inviting them to collaborate with Iranian intelligence."}],[{"start":91.65,"text":"Iran’s hackers have been experimenting with AI for years, but the powerful new models have made them more threatening. "}],[{"start":98.65,"text":"“This is all being done automatically,” said Gil Messing at Israeli cyber security firm Check Point. “They are using every tool they can in order to expedite their efforts through AI.”"}],[{"start":null,"text":"https://x.com/Trillion0x/status/2031911935671927231?s=20"}],[{"start":109.25,"text":"Iranian attacks often rely heavily on convincing unwitting targets to click dodgy links. But it can take weeks of chatting under a fake identity to build trust and coax them to let their guard slip. "}],[{"start":122,"text":"“If you are from Tehran and trying to pretend to be the HR person at a defence contractor, it is a heavy lift to talk to someone for a month and come off as a person living in California,” said the cyber security analyst. "}],[{"start":135.85,"text":"Shortly before the conflict began in late February, Google spotted state-backed group APT42 using the tech giant’s Gemini model to do just that."}],[{"start":145.75,"text":"Western companies have tried their best to keep the Iranians out, but spotting new accounts quickly becomes a game of whack-a-mole."}],[{"start":153.95,"text":"“Where we identify harmful activity, we take enforcement action, including disabling accounts, terminating access, or limiting capabilities being abused,” OpenAI said in a statement, adding that it had safeguards across its platform and that its most advanced models were “not broadly available for use”."}],[{"start":172.1,"text":"OpenAI said it regularly reported and disrupted attempts by Iran-affiliated actors to misuse its services, noting the actors generally used its models for activities including research, translation, debugging and scripting support. It said its models with safeguards offered no novel cyber capabilities."}],[{"start":189.9,"text":"Google declined to comment on the issue. "}],[{"start":192.9,"text":"Google found last year that Iranian hackers use its chatbot much more heavily than their more sophisticated North Korean, Russian or Chinese counterparts. APT42 had even been using it to study how to jam American F-35 fighter jets. "}],[{"start":209.45000000000002,"text":"But the Iranian efforts to integrate AI into the military are not limited to cyber operations. An FT analysis of some 300 Iranian military journal articles from the past five years found studies on AI bolstering electromagnetic warfare and accelerating decision-making in battlefield command centres, as well as improving drone guidance and underwater targeting.   "}],[{"start":232.95000000000002,"text":"“Iran wants to keep pace with the cutting-edge militaries,” said Nicole Grajewski, an expert on the Iranian military at Sciences Po."}],[{"start":null,"text":"

"}],[{"start":240.95000000000002,"text":"Mainstream western technological advances have helped Iran catch up before. With few speakers of English and Russian in Iran’s security establishment, Google Translate enabled them to study new western military research much faster, said Farzin Nadimi, an expert on the Iranian military at the Washington Institute for Near East Policy."}],[{"start":261.25,"text":"Washington’s own use of AI in its campaign against Iran helped it blitz through targets at a much higher pace than before, experts say. The US relies on Palantir’s Maven Smart System, a battlefield command dashboard, and Anthropic’s generative Claude model to interpret intelligence, build and suggest strikes, and provide real-time feedback during combat. "}],[{"start":284.2,"text":"Iran, whose economy and research institutions have been hobbled by sanctions, still lags far behind. "}],[{"start":290.95,"text":"But it has been creative. Some experts say Iran has been winning the online propaganda war, with state accounts regularly posting viral AI-generated videos mocking President Donald Trump."}],[{"start":303.45,"text":"It claims to have integrated AI into its missiles and drones. The Islamic Revolutionary Guard Corps (IRGC), for example, claims to have cruise missiles with AI guidance, navigation and electronic warfare evasion systems. "}],[{"start":318.5,"text":"While the Revolutionary Guards should be capable of using those technologies, it is unclear if they have successfully deployed them in combat, said Nadimi. "}],[{"start":327.3,"text":"But there are signs it may have used rudimentary AI in planning strikes across the Gulf, with the consistent targeting in its opening salvos potentially indicating that AI played a role, Grajewski said. "}],[{"start":338.7,"text":"Iran was likely also using AI for predictive analysis and to identify US force locations before firing on them, she added."}],[{"start":null,"text":""}],[{"start":346.25,"text":"Analysts said Iran was also almost certain to be using AI in ways that would be virtually impossible for outsiders to easily see. It has built much of its AI infrastructure on open-source and locally designed models, hosted away from prying eyes on closed networks. "}],[{"start":364.75,"text":"“What is being observed is only the tip of the iceberg,” said the cyber analyst."}],[{"start":370.35,"text":"Iran began experimenting with a national AI platform last year, designed to keep working even if the country disconnects itself from the wider global internet. It was developed by Tehran-based Sharif University, an institution under strict western sanctions for its close military ties."}],[{"start":387.1,"text":"“Investing in AI is really a national security modernisation programme,” said Alex Leslie, a senior adviser at cyber security firm Recorded Future who has studied Iranian AI use. The investment was also intended to help Iran overcome economic sanctions and mitigate wartime disruption. “It is a strategic imperative,” Leslie added. "}],[{"start":407.8,"text":"Israeli and US air strikes in early April severely damaged the data centre that hosts the core infrastructure for the platform, as well as laboratories involved in AI research at Sharif University. But it is unlikely that Israel and the US will be able to put the genie back in the bottle."}],[{"start":423.15000000000003,"text":"“This is a sophisticated nation with sophisticated tools,” said Check Point’s Messing. “They will be paying the price for whatever it is to get the best that they can.”"}],[{"start":438.45000000000005,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1780213789_4616.mp3"}