---
format: "aidr-story-markdown/v1"
id: "203cb704a0200cf0999b4e14a8905fc93509dadb6bc556f06d2513d4acfcb5f4"
canonical_url: "https://aidr.today/203cb704?lang=en"
title: "SPILLWAY: Separating Target Admission from Sanitization to Defend Tool-Using Agents against Prompt Injection"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-17T00:00:00.000Z"
category: "Research"
topics: ["safety"]
source_urls: ["https://huggingface.co/papers/2611.07222"]
summary: "Tool-using agents act on email, files, and payments, so an instruction hidden in a web page or a shared document can make an agent send data or money to an attacker. Many runtime defenses put a language model on the decision that lets an action through, so the decision depends on a model that may read the attacker's text and adds model calls to every step. Provenance defenses keep the model off the final check but need a model to plan or route each task first. Where an action goes can instead be read from the tools' typed returns: an email's sender or a payment's account arrives in a typed field, while injected instructions arrive as free text. SPILLWAY turns this observation into a deterministic floor with no per-task plan. The check admits a call when each target value that appears in tool output but not in the user's task was returned by a read-only tool in a typed field or is covered by a fixed delegation rule. A deterministic sanitizer removes instruction-like regions its fixed rules match before the model reads them and retains the removed spans as evidence, and a local judge may restore a refused call, never one whose target lies in the removed text."
---

# SPILLWAY: Separating Target Admission from Sanitization to Defend Tool\-Using Agents against Prompt Injection

> [Open the canonical story](<https://aidr.today/203cb704?lang=en>)

**Published:** 2026-11-17T00:00:00.000Z
**Category:** Research
**Topics:** safety

## Summary

Tool\-using agents act on email, files, and payments, so an instruction hidden in a web page or a shared document can make an agent send data or money to an attacker\. Many runtime defenses put a language model on the decision that lets an action through, so the decision depends on a model that may read the attacker's text and adds model calls to every step\. Provenance defenses keep the model off the final check but need a model to plan or route each task first\. Where an action goes can instead be read from the tools' typed returns: an email's sender or a payment's account arrives in a typed field, while injected instructions arrive as free text\. SPILLWAY turns this observation into a deterministic floor with no per\-task plan\. The check admits a call when each target value that appears in tool output but not in the user's task was returned by a read\-only tool in a typed field or is covered by a fixed delegation rule\. A deterministic sanitizer removes instruction\-like regions its fixed rules match before the model reads them and retains the removed spans as evidence, and a local judge may restore a refused call, never one whose target lies in the removed text\.

## Sources

- [Story source](<https://huggingface.co/papers/2611.07222>)

