Search papers, labs, and topics across Lattice.
1
0
2
5
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.
