// Security Advisory Triage
Command injection in AWS databases-on-aws agent plugin before 1.7.1
AWS security bulletins · published 2026-10-09 06:12 UTC
WHAT CHANGEDCVE-2026-107322 disclosed on 8 October 2026. The databases-on-aws plugin for AI coding agents, versions 1.0.0 through 1.7.0, has an incomplete input filter that allows a remote unauthenticated attacker to inject crafted content. If an agent ingests this content and invokes the local helper script, OS commands execute on the host with the helper process's permissions. Fixed in version 1.7.1.
WHY IT MATTERSAny system running an AI agent with this plugin installed can execute arbitrary commands if the agent processes attacker-controlled database content. The attack requires no authentication and no direct network access to the vulnerable code — only that the agent read malicious input from a source the attacker can influence. Commands run with whatever privileges the helper has, which in agent environments is often the developer's own account.
WHAT TO DOCheck whether databases-on-aws is in your agent dependencies or plugin directory. If you are running any version from 1.0.0 to 1.7.0, update to 1.7.1 immediately. If you cannot update, disable the databases-on-aws plugin until you can.
Model-reported confidence 95% (not measured accuracy)
Primary source ↗
written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
+## CVE-2026-107322 - OS command injection in Amazon Agent Plugins for AWS databases-on-aws +https://aws.amazon.com/security/security-bulletins/rss/2026-130-aws/ +Thu, 08 Oct 2026 19:29:19 +0000 + +Bulletin ID: 2026-130-AWS + +Scope: AWS + +Content Type: Important (requires attention) + +Publication Date: 10/08/2026 11:30 AM PDT +Description: +Amazon Agent Plugins for AWS is an open source collection of plugins that extends supported AI coding agents with AWS-focused workflows and tool integrations. The databases-on-aws plugin provides database design, development, migration, and operational guidance, including Aurora DSQL helper scripts. +We identified CVE-2026-107322, where an incomplete list of disallowed inputs in databases-on-aws versions before 1.7.1 could allow a remote unauthenticated actor to supply crafted content that an agent ingests. The affected code is not a listening network service; operating-system command execution on the host is possible only if the agent subsequently invokes the local helper with the crafted database command value. Any resulting command runs with the permissions of the process running the helper. This issue does not affect the Aurora DSQL service or bypass database privileges. +Impacted versions: databases-on-aws versions 1.0.0 through 1.7.0 +Please refer to the article below for the most up-to-date and complete information related to this AWS Security Bulletin.
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
| triage verdict | MATERIAL at 0.92, needed 0.55 |
|---|---|
| triage reasoning | OS command injection in AWS AI agent plugin affecting teams running AI coding agents with database workflows. |
| signals | known vulnerability in AI agent plugin · OS command injection via crafted input · affects AI SDK/agent framework (in-scope runtime) · remote unauthenticated attack vector · fixed version available (1.7.1) · open source plugin likely in dependency tree of teams using AWS agents |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |