Adversarial Misuse of Generative AI

Google’s GTIG analyzed government-backed actors’ use of the Gemini LLM and found that actors—particularly Iranian APT and IO groups—used Gemini for routine operational tasks (research, code troubleshooting, content creation, localization, target reconnaissance) but did not demonstrate novel AI-enabled attack techniques; publicly available jailbreak prompts were attempted but generally blocked by Gemini’s safety controls.