AI Paper Atlas

Research map / Large language models / Agents, code and math

PACEbench: A Framework for Evaluating Practical AI Cyber-Exploitation Capabilities

Zicheng Liu, Lige Huang, Jie Zhang, Dongrui Liu, Yuan Tian, Jing Shao

ICLR 2026 · Large language models › Agents, code and math · approach: Software engineering agents · 10 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
use PACEagent instead of CAI agent frameworkClaude-3.7-Sonnet · PACEbench A-CVEscore: +0.18Improved
use PACEagent instead of CAI agent frameworkClaude-3.7-Sonnet · PACEbench B-CVEscore: +0.05Improved
use PACEagent instead of CAI agent frameworkClaude-3.7-Sonnet · PACEbench C-CVEscore: +0.20Improved
use PACEagent instead of CAI agent frameworkClaude-3.7-Sonnet · PACEbench overallscore: 65.2% performance gainImproved
PACEagent multi-stage design vs CAI token useClaude-3.7-Sonnet · PACEbenchtoken consumption: 28% more tokensHurt
add benign hosts to environment (B-CVE vs A-CVE)PACEagent · PACEbenchscore: Claude-3.7-Sonnet 0.412 -> 0.263Hurt
add multi-host dependencies (C-CVE vs B-CVE)PACEagent · PACEbenchscore: Claude-3.7-Sonnet 0.263 -> 0.267Hurt
add WAF defenses (D-CVE)PACEagent · PACEbenchscore: 0.241 -> 0.000Hurt
increase vulnerability difficulty (lower human pass rate)PACEagent · PACEbench A-CVEnumber of successful modelsHurt
use open-source models with small context windows instead of proprietary onesPACEagent · PACEbenchscore: Qwen3-32B 0.024, Deepseek-V3 0.012 vs Claude-3.7-Sonnet 0.241Hurt
use Claude 4 instead of Claude 3.7PACEagent · PACEbench preliminarysuccess rateHurt

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