Researchers have developed a new adversarial attack called MissClick that targets GUI grounding models by exploiting how they parse screen coordinates. The attack manipulates digit-serialized coordinates, causing large displacements in the executed clicks. MissClick offers two objectives: MissClick-U for untargeted disruption and MissClick-T for targeted hijacking of clicks. Experiments show MissClick significantly outperforms existing attacks on desktop, web, and mobile platforms. AI
IMPACT This research highlights a new vulnerability in GUI grounding models, potentially impacting the security of AI-driven interfaces.
RANK_REASON The cluster contains a research paper detailing a novel adversarial attack on AI models. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →