On February 15, 2026, reports emerged indicating that Google had recently revealed its premier AI model, Gemini, is currently the target of extensive, commercially motivated distillation attacks. In these attacks, adversaries are methodically and persistently bombarding the model with meticulously designed prompts—exceeding 100,000 attempts per attack—in a bid to reverse-engineer Gemini's internal reasoning processes and decision-making frameworks. The ultimate aim is to clone the model or bolster their own AI capabilities.
These assaults are predominantly launched by private AI firms or research organizations striving for a competitive advantage, with the origins of the attacks tracing back to various global regions. Google has underscored that such distillation attacks amount to intellectual property theft. Despite the implementation of detection and blocking measures, the inherent openness of mainstream large-scale model services continues to render them susceptible to such threats.