# Preprint reports blocking hidden-image prompt attacks in MCP workflows

_Published Monday, October 26, 2026 at 7:56 AM EDT · Science, Security · Latest · Tier 2 — Notable_

Researchers report that a defense blocked all 24 adaptive attack payloads in a Model Context Protocol (MCP) testbed, where tools connect AI agents to external information. The preprint tested instructions hidden in email-signature images using three concealment methods. The undefended pipeline had 100% attack success; keyword and regular-expression defenses each allowed 62.5% of attacks.

The defense authenticates trust labels, tracks where decoded content originated, and blocks critical tool calls influenced by untrusted material. An audited process for promoting trusted content raised legitimate-task completion from 67% to 100%. The containment depends on fully mediated MCP execution environments.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2610.31974)

---
Canonical: https://techandbusiness.org/newswire/1Dh11b48Vfc1mCMqAzuFnw
Published: 2026-10-26T11:56:25.676Z
Story chronology: 2026-10-26T04:00:00.000Z
Retrieved: 2026-10-26T14:49:14.678Z
Publisher: Tech & Business (techandbusiness.org)