# Paper Pushes Agent Safety From Prompt Filtering to Fail-Closed Control of Tool Actions

The proposal wraps agent tool calls in action boundaries, trusted provenance and execution that stops by default when a check fails.

- Published: 2026-08-19T06:20:14.557Z
- Canonical: https://polylog.news/ai/2026-08-19/paper-pushes-agent-safety-from-prompt-filtering-to-fail-clos
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.AI](https://arxiv.org/abs/2608.16891)

A preprint on runtime governance for agentic AI begins with an observation that anyone running AI agents in production has likely already made. When a system can modify files, send messages, launch jobs and change the state of a workflow, t…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-08-19/paper-pushes-agent-safety-from-prompt-filtering-to-fail-clos (subscription information: https://polylog.news/pricing).