Research map / Large language models / Agents, code and math
AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluation
Changyi Li, Pengfei Lu, Xudong Pan, Fazl Barez, Min Yang
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 |
|---|---|---|---|
| add high stress and high temptation together to the environment | frontier LLM agents · X-Bench average | risk rate (%): 21.7 -> 54.5 | Hurt |
| add coupled stress and temptation | Qwen3-32B · X-Bench | risk rate (%): 41.3 -> 53.2 | Hurt |
| move from baseline to S1T1 | Gemini-3-Flash · X-Bench | risk rate (%): 25.0 -> 70.6 | Hurt |
| move from baseline to S1T1 | Gemini-2.5-Flash · X-Bench | risk rate (%): 26.5 -> 52.2 | Hurt |
| add high temptation only | average over 9 models · X-Bench | risk rate (%): 21.7 -> 32.8 | Hurt |
| add high stress only | average over 9 models · X-Bench | risk rate (%): 21.7 -> 29.7 | Hurt |
| scale Qwen3 from 4B to 32B | Qwen3 · X-Bench Capability Misuse | risk rate (%): 65.0 -> 48.6 | Improved |
| scale Qwen3 from 4B to 32B | Qwen3 · X-Bench Specification Gaming | risk rate (%): 28.6 -> 43.8 | Hurt |
| upgrade Gemini-2.5-Flash to Gemini-3-Flash | Gemini · X-Bench Specification Gaming | risk rate (%): 26.2 -> 53.5 | Hurt |
| replace executable-code logic layer with text-as-state simulation (Petri) | AutoControl Arena · Human fidelity comparison | preference rate (%): 60% | Improved |
| add dual-loop verification (semantic and syntax correction) (ablation) | Coder agent · X-Bench generation | end-to-end success rate: over 98%, 69 of 70, 2.8 cycles | Improved |
Closest papers by meaning
- GT-HarmBench: Benchmarking AI Safety Risks Through the Lens of Game Theory (NeurIPS 2026 · similarity 0.87)
- Claw-Eval: Towards Trustworthy Evaluation of Autonomous Agents (NeurIPS 2026 · similarity 0.87)
- AgentAuditor: Human-level Safety and Security Evaluation for LLM Agents (NeurIPS 2025 · similarity 0.87)
- AGENTSAFE: Benchmarking the Safety of Embodied Agents on Hazardous Instructions (CVPR 2026 · similarity 0.87)
- Building a Foundational Guardrail for General Agentic Systems via Synthetic Data (ICLR 2026 · similarity 0.87)
- PACE: A Proxy for Agentic Capability Evaluation (NeurIPS 2026 · similarity 0.87)
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.
