Translation result.
[Herald Economy=Reporter Doh Hyun-jung] Iran has maintained a cease-fire with the U.S., but its cyber campaign has accelerated. Iranian operators are increasingly exploiting AI models from Western tech firms, including ChatGPT and Gemini.
The Financial Times reported on the 31st of last month (local time) that Tehran used ChatGPT and Gemini to develop malicious code and to craft phishing messages in Hebrew and Arabic, leveraging Western AI models to sustain its campaign.
Iran’s operations often rely on luring targets to click malicious links. FT analysts say recent AI advances enable operators to create convincing fake personas that deceive targets in the U.S. and Israel.
Tehran also uses social media to distribute posts and videos mocking President Donald Trump — a form of information warfare. AI models are used to produce and amplify that content.
Despite the cease-fire, Iran’s cyber operations have persisted and have hit U.S. allies hard. The UAE has reportedly been targeted with more than 500,000 cyber incidents daily tied to attacks leveraging ChatGPT. Israel has been inundated with phishing emails and texts aimed at coaxing targets into cooperation with Iranian intelligence.
Google said it detected in February — just ahead of the recent hostilities — that the government-backed hacking group APT42 used Gemini for cyberwarfare. Google also found last year that Iranian actors used Gemini far more than counterparts in North Korea, Russia or China. Researchers say APT42 even used Gemini to study techniques for jamming U.S. F-35 fighter jets.
The FT warns that Iranian hackers, who have experimented with AI for years, have grown more capable and dangerous as they adopt more powerful models.
OpenAI, maker of ChatGPT, said it found Iranian operators used ChatGPT for research, translation, debugging and scripting support. The company said it routinely reports and blocks abuse, disabling accounts, cutting access and restricting features when it detects harmful activity. It also maintains tighter controls on the most advanced models.
Most Commented