伊朗正把ChatGPT、Gemini等西方AI模型用于网络战：它们可协助开发恶意软件、撰写波斯文外的钓鱼信息、生成假身份，并以更高速度与规模发动攻击。安全分析人士称，AI已贯穿其行动全流程；阿联酋称每日遭受逾50万次网络攻击，部分与OpenAI的ChatGPT有关。

FT对近5年约300篇伊朗军事期刊文章的分析显示，AI被设想用于电磁战、指挥中心决策加速、无人机导航和水下目标识别；伊朗革命卫队也声称其巡航导弹具备AI制导、导航和电子战规避系统。尽管伊朗经济与科研受制裁拖累，其仍通过开源和本地模型、封闭网络与国防体系推进AI国家平台，力图在遭遇断网时继续运行。

OpenAI与Google都曾发现伊朗相关账号滥用其模型，OpenAI称会禁用账号、终止访问或限制能力，但伊朗仍可借AI进行研究、翻译、调试、脚本支持，以及面向美国和以色列的心理战与情报诱骗。文章同时指出，美国在对伊行动中也使用Palantir的Maven和Anthropic的Claude提升打击速度；今年4月，美以空袭重创了伊朗AI平台核心数据中心与Sharif University实验室，但外界认为伊朗的AI能力难以被彻底清除。

Iran is using ChatGPT, Gemini and other Western AI models to scale cyber operations, generating malware, phishing lures and fake personas while automating much of the workflow, according to cyber experts and firms. The UAE said it faced more than 500,000 cyberattacks per day, and Israeli users were hit by waves of phishing messages tied to Iran-linked activity. (Key numbers: 50)

An FT review of about 300 Iranian military journal articles from the past five years found research on AI for electromagnetic warfare, faster battlefield command decisions, drone guidance and underwater targeting. Iran’s military also claims AI-enabled cruise missiles, though analysts say combat deployment is unproven; there are signs AI may have helped with strike planning and predictive analysis against US force locations in the Gulf.

Despite sanctions and damage from April airstrikes on Sharif University and a key data center, Iran is building a national AI platform meant to work even if disconnected from the global internet. OpenAI and Google have repeatedly disrupted Iranian misuse, but analysts say much of Iran’s AI infrastructure sits on open-source and locally designed models, meaning what is visible is only the tip of the iceberg. (Key numbers: 4)