Research map / Large language models / Agents, code and math
OTora: A Unified Red Teaming Framework for Reasoning-Level Denial-of-Service in LLM Agents
Xinyu Li, Ronghui Mu, Lin Li, Tianjin Huang, Gaojie Jin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| insert trigger at best-scored position instead of random valid position (ablation) | OTora Stage I · WebShop, LLaMA-3.1-70B | ASR_H: 55% -> 74% | Improved |
| best-position insertion instead of fixed prefix (ablation) | OTora Stage I · WebShop, LLaMA-3.1-70B | ASR_H: 49% -> 74% | Improved |
| add fidelity term to RTI-only payload scoring (ablation) | OTora Stage II · WebShop, LLaMA-3.1-70B | Hit rate: 52% -> 74% | Improved |
| add stability term to RTI+fidelity scoring (ablation) | OTora Stage II · WebShop, LLaMA-3.1-70B | RTI: 8.9x -> 10.8x | Improved |
| add attention-aware term to insertion scoring (ablation) | OTora Stage I · WebShop, LLaMA-3.1-70B white-box | ASR_H: 73% -> 82% | Improved |
| add dynamic target co-evolution instead of fixed target (ablation) | OTora Stage I · WebShop, LLaMA-3.1-70B white-box | ASR_H: 84% -> 90% | Improved |
| environment-level instead of instruction-level injection | OTora Stage I · WebShop/Email/OS agents | ASR: 3--6% on average | Improved |
| OTora-Persistent vs OTora-ICL(Agnostic) payload | OTora Stage II · WebShop, LLaMA-70B | Hit rate: 60% -> 92% | Improved |
| OTora trigger optimizer vs UDora | Stage I white-box trigger · WebShop, LLaMA-3.1-70B | ASR_H: 94% -> 95% | Improved |
| larger backbone under same sink payload | OTora Stage II · WebShop | RTI: 5.1x -> 9.7x | Improved |
Closest papers by meaning
- UDora: A Unified Red Teaming Framework against LLM Agents by Dynamically Hijacking Their Own Reasoning (ICML 2025 · similarity 0.90)
- Co-RedTeam: Orchestrated Security Discovery and Exploitation with LLM Agents (ICML 2026 · similarity 0.88)
- AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases (NeurIPS 2024 · similarity 0.87)
- LITMUS: Benchmarking Behavioral Jailbreaks of LLM Agents in Real OS Environments (NeurIPS 2026 · similarity 0.86)
- Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents (ICLR 2025 · similarity 0.86)
- Metis: Learning to Jailbreak LLMs via Self-Evolving Metacognitive Policy Optimization (ICML 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
