Search papers, labs, and topics across Lattice.
The University of Texas at Arlington
3
0
4
Over 65% of victim queries in online searches lead to potentially harmful links, exposing critical flaws in digital support systems for technology-facilitated abuse victims.
Poisoned data can alter summarization behavior without triggering alarms, but our defense framework detects and restores model integrity with impressive precision.
Jointly verifying user intent and response harm can reduce attack success rates to just 4.1%, setting a new standard for LLM safety.
