VeriHarm-Bench: Measuring Harmful Success in LLM Coding Agents Verifiably
Factual evidence
What the source reports
Researchers introduced VeriHarm-Bench, a benchmark measuring collateral system damage caused by LLM coding agents executing benign shell commands.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.CL — Computation and Language · RESEARCH
- Published by source
- 4 November 2026
- Collected by OneBench
- 5 Nov 2026, 02:01 UK
Stored source excerpt
arXiv:2611.03575v1 Announce Type: new Abstract: Large language model (LLM) coding agents increasingly operate with shell access, allowing them to write scripts, move files, alter permissions,…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Autonomous coding agents executing benign tasks can silently corrupt file permissions and infrastructure controls, introducing unmonitored operational risk.
Do what
Review execution sandboxing and file-system control boundaries with the engineering team before deploying shell-capable coding agents in production environment.