University of Southern California

Headlines · 2

1. 3d
   Agent MechSuits Adds Internal Safety Controls to CLI Agents

   Researchers from Georgia Tech, Rutgers, the University of Southern California and other institutions proposed Agent MechSuits, a NeurIPS 2026-accepted method that reads hidden-state safety signals before each command-line agent action and intervenes on 10 dimensions without updating the main model. Across LLaMA-3.1-8B, Qwen-2.5-7B and Gemma-2-9B, the reported MAS current-step detection accuracy averaged 97.32%, while Qwen-2.5-7B's malicious-step ASR fell from 91.30% to 4.35%. The method also had limitations: generation-collapse rate rose from 2.24% to 18.54%, the evaluation used language-model judging, and the probe did not reliably detect prompt injection arriving through environment feedback.

   新智元 · 🔥 0
2. Oct 31
   CLASHarness Benchmarks Multimodal Agents in Multi-Agent Conflicts

   A research consortium involving Google and six universities released CLASHarness, an open benchmark for testing 15 models when teammate reports conflict with system evidence. According to New Zhiyuan, its 600 executable tasks use high-resolution visual evidence and cover 20 domains; the report says no model exceeded a 30% end-to-end pass rate, with active verification and task handoff remaining major weaknesses. The generation pipeline, evaluation tasks, and an interactive trajectory viewer are available as open-source materials.

   新智元 · 🔥 0

Experience and discussion from the community

Share my University of Southern California experienceAsk about University of Southern California