Before Retrieval Begins: Safety and Tool-Use Decisions in LLM Agents
Factual evidence
What the source reports
A paper finds seven open-weight models execute harmful requests via tool calls 74.5% of the time without retrieval safety checks.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.CL — Computation and Language · RESEARCH
- Published by source
- 19 October 2026
- Collected by OneBench
- 20 Oct 2026, 03:01 UK
Stored source excerpt
arXiv:2610.23071v1 Announce Type: new Abstract: Large language models increasingly act as agents that call external tools, yet most tool-safety research examines only what follows a…
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
Agentic workflows invoking external tools present unmitigated security and safety risks prior to standard retrieval-guardrail execution.
Do what
Review pre-execution tool-calling safety policies with model risk management teams when deploying open-weight agentic models.