Attackers Move From Chatting With AI to Letting It Run the Attack
Google Threat Intelligence Group (GTIG) says forward-leaning attackers have moved beyond simple AI prompting to agentic AI workflows and automation. This cuts the time humans spend directing an attack, and shrinks the window defenders have to respond. In Q2 2026, GTIG saw threat actors compromise a cloud resource, then plan, build and run an agent-enabled mass credential harvesting campaign in under six hours. GTIG also tracked a group called UNC6780 using several tactics to trick AI coding assistants and large language model (LLM) security scanners as part of open source software supply chain compromises.
AI assets are becoming targets too. GTIG observed attackers going after proprietary AI models, source code, prompts and research in sectors including healthcare, government and media. Others stole API credentials or took over victim cloud environments to run unauthorised AI workloads. GTIG notes that everything from model weights to cloud compute quotas is now valuable for espionage, extortion and resource theft.
State-sponsored and cyber crime groups also continue to use AI as a force multiplier across the attack lifecycle. This includes target reconnaissance, writing convincing social engineering lures, disguising custom malware and troubleshooting after a breach. Some are experimenting with scaling information operations campaigns. Note that this article is based on the opening of GTIG's report, not the full findings.