# Researchers test defense against attacks carried between AI agent sessions

_Published Monday, October 12, 2026 at 1:25 PM EDT · Science, Security · Latest · Tier 2 — Notable_

Researchers report in a preprint that a runtime defense reduced attacks carried between AI agent sessions through persistent files, configurations and cached data. Their SkillRelayBench benchmark contains 270 adversarial cases across 30 scenarios, testing how artifacts created by one skill can influence later executions.

The defense, SkillRelayGuard, restores artifact provenance and reviews actions before execution. Across three agent settings, it reduced attack success from 69.5% to 10.9% and attack chain success from 54.8% to 10.0%, while legitimate task completion rose from 79.6% to 87.5%. The agent retains the final execution decision.

## Sources

- [cs.CR updates on arXiv.org](https://arxiv.org/abs/2610.13227)

---
Canonical: https://techandbusiness.org/newswire/Zmu1hkVAwg66IzLdp_RVqb
Published: 2026-10-12T17:25:32.218Z
Story chronology: 2026-10-12T04:00:00.000Z
Retrieved: 2026-10-12T19:52:13.016Z
Publisher: Tech & Business (techandbusiness.org)