SMSMC Blog

AI developer tools are on the patch list now: AWS's November bulletins

In the first week of November 2026, AWS published fixes for an MCP server, an AI workflow tool and a machine learning library. Why AI tooling belongs in your inventory.

By Jim Venuto

Between November 2 and November 5, 2026, AWS published security bulletins for several open-source tools that developers use to build with AI. Three stand out:

  • AWS Serverless MCP Server (CVE-2026-116635). This is a Model Context Protocol server that lets an AI assistant work with serverless applications on AWS. In versions before 0.2.1, AWS says a flaw “might allow an actor who can influence the MCP agent to read the contents of JSON files outside the intended deployment status directory,” and, where write access is enabled, create or overwrite JSON files outside it. AWS lists no workaround; the fix is to upgrade.
  • aidlc-traceability in AWS AI-DLC Workflows (CVE-2026-120497). A stored cross-site scripting flaw in the tool’s HTML report generator. AWS recommends removing any local copy obtained before July 3, 2026, or updating it, and notes the tool was removed from its original repository on July 16, 2026.
  • GluonTS (CVE-2026-119083). A time-series forecasting library. Before version 0.17.0, a crafted archive could write files outside the folder it was extracted to, with the privileges of the process doing the extraction.

AWS does not say any of these were exploited. For each, AWS recommends upgrading or replacing the affected copy and making sure “any forked or derivative code” picks up the fix.

Why these matter more than their severity

None of these is the kind of flaw that makes headlines. What makes them worth a small organization’s attention is where they run.

AI developer tools usually run on a developer’s own machine, installed by that developer, often with the developer’s own cloud credentials. AWS says its MCP servers are designed to support standard input and output only, which means each one runs as a local process beside the AI assistant, typically on a developer’s machine and with that machine’s cloud access. The flaw in the Serverless MCP Server is reachable by anyone who can “influence the MCP agent,” which, for an AI assistant reading documents, code or web pages, can be content the developer did not write.

These tools also do not appear in the places most organizations look for software. They are not in the endpoint management console, the asset list or the vendor register. They come from a package manager, get updated whenever the developer remembers, and are copied into other projects, which is why AWS keeps mentioning forks.

We covered how MCP servers widen the attack surface in MCP Server Security. This week’s bulletins are the maintenance side of the same problem.

Pointers for AI tooling

  1. Ask your developers and contractors what AI tools they run. MCP servers, coding assistants, AI workflow tools and machine learning libraries. Start with a simple list and an owner for each.
  2. Check versions against this week’s fixes. The AWS Serverless MCP Server should be 0.2.1 or later, GluonTS 0.17.0 or later, and old local copies of aidlc-traceability should be removed or updated.
  3. Give AI tools their own, narrower credentials. An MCP server that only needs to read deployment status should not run with an administrator’s cloud profile. Write access should be off unless it is needed.
  4. Limit what a local AI tool can reach. The MCP project’s own security guidance recommends local servers use the standard input and output transport to limit access to the client, and that servers never accept tokens that were not issued for them.
  5. Subscribe to the bulletins for the platforms you use. AWS, Microsoft and Google each publish security bulletins; someone should read them weekly.

Who does what

Leadership decides which AI development tools are approved and requires that they be on the inventory. Developers and contractors keep the list current, update tools promptly and use the narrow credentials. Your IT team or MSP watches the vendor bulletins and checks versions during routine reviews.

Where outside help fits

AI tooling arrives faster than any approval process designed for traditional software. SMC GRC.ai brings it into governance: a shadow AI inventory that includes developer tools, an AI use policy with an approval path, and AI vendor risk review. CyberIntelPro sets the architecture for how those tools authenticate and what they may reach in your cloud accounts, and gives your team the change list.

Want to talk it through? Book a free 30-minute call with SMC GRC.ai.

Sources