CVE-2026-26030: Microsoft Semantic Kernel InMemoryVectorStore filter functionality vulnerable to remote code execution
The Semantic Kernel Python SDK's `InMemoryVectorStore` filter feature allows code injection, which gives an authenticated attacker remote code execution. It affects the `semantic-kernel` package before 1.39.4. Upgrade to 1.39.4, or avoid `InMemoryVectorStore` in production.
Published Feb 19, 2026 · Updated Jun 8, 2026 · Analysis Oct 31, 2026
What an attacker can do
InMemoryVectorStore lets callers pass a filter expression to select records. The fix, pull request 13505, blocks dangerous attribute names that must not be accessed in filter expressions. That tells us the filter text was evaluated in a way that allowed an attacker to reach Python internals through attribute access. The advisory classes it as code injection (CWE-94) and does not publish an exploit walkthrough.
The CVSS vector says network attack, low privileges, no user interaction and changed scope, with high confidentiality, integrity and availability impact. In practice, an attacker who can influence the filter string reaches code execution in the process hosting the SDK. In an agent application, that means whoever can shape a search filter, whether through an API field or text an LLM turns into a filter, may run code with the app's access to model keys, data and tools.
The sources do not say which input paths are exposed in typical apps, so check where your code builds filters from outside input. EPSS places it in the 91st percentile for this batch.
Who is affected
semantic-kernel on PyPI below 1.39.4 is affected, for applications using InMemoryVectorStore with filters. Check your version:
pip show semantic-kernelSearch your code for InMemoryVectorStore and for any place where a filter is built from user or model-generated input.
semantic-kernel
- Ecosystem
- PyPI
- Affected
- < 1.39.4
- Fixed in
- 1.39.4
- Check
pip show semantic-kernel
How to fix it
Upgrade to semantic-kernel 1.39.4, released as python-1.39.4.
pip install -U "semantic-kernel>=1.39.4"The advisory's workaround is to avoid InMemoryVectorStore in production scenarios. Do not pass untrusted text into filter expressions while you are on an older version.
How to tell if you were hit
Review application logs for unusual filter strings sent to vector store queries, especially ones containing double-underscore attribute names such as __class__ or __globals__. On the host, look for unexpected child processes, new files and outbound connections from the Python process that runs the SDK.
Original advisory
From the GitHub Advisory Database (CC BY 4.0).
Impact:
An RCE vulnerability has been identified in Microsoft Semantic Kernel Python SDK, specifically within the InMemoryVectorStore filter functionality.
Patches:
The problem has been fixed in python-1.39.4. Users should upgrade this version or higher.
Workarounds:
Avoid using InMemoryVectorStore for production scenarios.
References:
Release python-1.39.4 · microsoft/semantic-kernel · GitHub PR to block use of dangerous attribute names that must not be accessed in filter expressions
Credited to amiteliahu, doredry, urioren.
Questions
- Does CVE-2026-26030 affect other vector stores?
The advisory names only the
InMemoryVectorStorefilter functionality in the Python SDK. It does not say other stores are affected.- Is there a workaround?
Yes: do not use
InMemoryVectorStorefor production. Upgrading to 1.39.4 is the actual fix.- Is it exploited in the wild?
No source reports exploitation, and it is not in CISA's KEV catalog.
References
Read for this analysis:
- https://nvd.nist.gov/vuln/detail/CVE-2026-26030
- https://github.com/microsoft/semantic-kernel/pull/13505
- https://github.com/microsoft/semantic-kernel/releases/tag/python-1.39.4
- https://github.com/pypa/advisory-database/tree/main/vulns/semantic-kernel/PYSEC-2026-163.yaml
- https://github.com/advisories/GHSA-xjw9-4gw8-4rqx
How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology
