← all datasets

ClawTrojan

Emerging
1papers using it
2026first seen

ClawTrojan is a benchmark designed to identify multi-step trojan attacks in local agentic harnesses, containing scenarios that demonstrate how attackers can embed prompt injections within files or tool outputs to achieve persistent control over LLM agents.

Papers using ClawTrojan (1)

ClawTrojan β€” datasets β€” cybersecurity