Search papers, labs, and topics across Lattice.
N\mathcal{Q}=\{q_{i}\}_{i=1}^{N} related to global public issues, spanning 18 languages 鈩抃mathcal{L}, from online platforms (e.g.e.g., Reddit), along with their corresponding answers and responses (named Answer) when available. We further supplement missing Answers by leveraging multiple LLMs. Specifically, for each Question qi鈭堭潚琿_{i}\in\mathcal{Q} we obtain two distinct answers: (1) a normal one ainorma_{i}^{\text{norm}}, either sourced directly from online platforms or generated by safety-aligned LLMs (e.g.e.g., GPT-5.2), tends to reflect socially accepted values; and (2) a risky one airiska_{i}^{\text{risk}}, generated by an uncensored version of open-source LLMs111https://huggingface.co/huihui-ai/models, Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security
1
0
2
7
By pinpointing the causal origins of tool use, AttriGuard neutralizes indirect prompt injection attacks that can hijack LLM agents, even when faced with adversarial optimization.
