AI Paper Atlas

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

ICML 2026 · Large language models › Agents, code and math · approach: Long-horizon LLM agents · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add high stress and high temptation together to the environmentfrontier LLM agents · X-Bench averagerisk rate (%): 21.7 -> 54.5Hurt
add coupled stress and temptationQwen3-32B · X-Benchrisk rate (%): 41.3 -> 53.2Hurt
move from baseline to S1T1Gemini-3-Flash · X-Benchrisk rate (%): 25.0 -> 70.6Hurt
move from baseline to S1T1Gemini-2.5-Flash · X-Benchrisk rate (%): 26.5 -> 52.2Hurt
add high temptation onlyaverage over 9 models · X-Benchrisk rate (%): 21.7 -> 32.8Hurt
add high stress onlyaverage over 9 models · X-Benchrisk rate (%): 21.7 -> 29.7Hurt
scale Qwen3 from 4B to 32BQwen3 · X-Bench Capability Misuserisk rate (%): 65.0 -> 48.6Improved
scale Qwen3 from 4B to 32BQwen3 · X-Bench Specification Gamingrisk rate (%): 28.6 -> 43.8Hurt
upgrade Gemini-2.5-Flash to Gemini-3-FlashGemini · X-Bench Specification Gamingrisk rate (%): 26.2 -> 53.5Hurt
replace executable-code logic layer with text-as-state simulation (Petri)AutoControl Arena · Human fidelity comparisonpreference rate (%): 60%Improved
add dual-loop verification (semantic and syntax correction) (ablation)Coder agent · X-Bench generationend-to-end success rate: over 98%, 69 of 70, 2.8 cyclesImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy