Decompose LLM vulnerability discovery into sub-tasks
Break down LLM vulnerability discovery into distinct, measurable steps for better evaluation.
Vulnerability discovery is a composite ability that can be decomposed into distinct sub-tasks for more precise measurement.
Instead of a single end-to-end verdict, evaluating an LLM's ability to comprehend source code, infer input constraints, construct inputs, execute them, and iteratively correct from feedback allows for localization of failure points and more targeted improvements.
The SWE-Test benchmark decomposes discovery into Open-loop (input construction without feedback), Feedback-enabled (input construction with a distance oracle), and Online Arena (path exploration by coverage gain) modes.
Evaluate LLM vulnerability discovery by breaking it into constituent parts: code comprehension, input constraint inference, input construction, execution, and feedback-driven correction. This allows for more granular assessment and targeted improvement of LLM security agents.
This decomposition helps isolate specific weaknesses in an LLM's reasoning or generation process, moving beyond a simple pass/fail to understand where the model struggles.
- SWE-Test: Benchmarking LLM Vulnerability Discovery via Input Prediction ↗arXiv cs.AI · arXiv cs.AI · Sep 5, 2026 · paraphrased
First seen on Subvolts Sep 5, 2026. Updated Sep 11, 2026. Spotted a problem? Tell us.
- SWE-Test: Benchmarking LLM Vulnerability Discovery via Input PredictionAnthropic · arXiv cs.AI · Sep 5, 2026
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Anti-patternRelying on gameable benchmarks for LLM vulnerability discoveryAvoid LLM benchmarks that are susceptible to data contamination or score recall against unknowable sets. · 5 min
- HeuristicAuxiliary environment signals improve LLM authorization accuracyProviding LLMs with auxiliary environment signals can substantially improve their accuracy in authorization decisions. · 30 min
- Case studyUsing LLMs to build tools for iOS location data analysisResearchers used an LLM to build a tool that decoded iOS location data, overcoming LLM analysis limitations. · 60 min
- HeuristicIterative refinement for data science agent planningRefine data science agent plans iteratively using a judge and router to improve solutions and handle complexity. · 5 min
- HeuristicLLMs are unreliable for direct privacy analysis but good toolmakersUse LLMs to build analysis tools, not for direct forensic or privacy analysis. · 15 min
- ToolMoMHa for multi-objective LLM harness optimizationOptimizes LLM harnesses for accuracy, safety, and token cost using a joint-reward proposer. · 120 min
Questions people ask
- Which AI tools does this work with?
- Any current model. It was seen with in the sources.
- Where did this come from?
- arXiv cs.AI (arXiv cs.AI). Subvolts collected, normalized and credited it; the link above opens the original.