Search papers, labs, and topics across Lattice.
Affiliation:
5
2
6
7
JustMem is introduced, which stores conversation history as compact atomic memories and adapts memory access along two dimensions to each query and achieves the highest mean accuracy and retrieval recall among the compared memory systems while using substantially fewer generative-model tokens for memory construction and inference.
PIMiner achieves impressive attack success rates against multiple LLMs with minimal query requirements, revolutionizing prompt injection red teaming.
Red-teaming long-context LLMs just got a whole lot cheaper: FlashRT slashes the compute and memory costs of prompt injection attacks by up to 7x.
Turns out, state-of-the-art prompt injection defenses aren't as robust as we thought: they crumble against adaptive attacks and struggle when the injected task aligns with the intended one.
AgentWatcher spots prompt injections in long-context LLMs by pinpointing the few key text snippets that actually influenced the model's behavior, then checking those against a clear rulebook of forbidden commands.
