CORTEXA
← Browse

Devina Jain

1 paper indexed

arxivcs.CRcs.AIcs.LG2026-07-20

Adaptive Adversaries: A Multi-Turn, Multi-LLM Benchmark for LLM Agent Security

Devina Jain, David Hartmann, Chuan Li

LLM-based agents process external content, exposing them to prompt injection and multi-turn manipulation. Most safety benchmarks evaluate defenders against fixed attack pools collected before evaluation, single-turn or multi-turn. We present a 21-scenario benchmark for \emph{adap…

View free PDFSource page