Google GTIG Traces UNC6780 AI Pipeline Building Credential Harvester in Under Six Hours

GTIG reporting shows the actor embedded at least seven distinct AI-assisted techniques inside its Dustmaker stealer and released Shai-Hulud and Miasma as open source. These releases lower the bar for emulation by groups previously limited to manual scripting. The same pattern appears in UNC6508 operations against North American research institutions and in Basin Castle’s LLM queries for target profiling and lure translation. Contract awards and procurement records indicate multiple PRC and Iranian actors are now testing agentic exploitation pipelines; GTIG’s disruption logs confirm Gemini accounts tied to Ravine Castle and Calanque Ion were disabled after documented misuse across the full attack lifecycle. Official attribution statements list state sponsorship, yet the technical artifacts—prompt logs, generated code commits, and model-extraction attempts—remain the only verifiable evidence. Open-source supply-chain compromise velocity has increased because AI reduces the skilled labor required for each stage. Defenders relying on model hardening alone create new surfaces for distillation attacks. GTIG’s real-time blocking of adversarial projects is reactive; the next measurable signal will be whether three or more independent actors replicate the six-hour campaign template within 90 days.