Skip to content
Agent Radar

Semantic Kernel 1.46.0 hardens plugins and connectors

Updated Semantic Kernel

Summary

Semantic Kernel 1.46.0 for Python (Oct 28, 2026) blocks ../ upload paths, pins HttpPlugin DNS, stops logging argument values and fixes connectors.

Tool
Semantic Kernel
Category
Frameworks
Date
Open source
Yes
MCP
Yes
Pricing
Not stated in the sources we cited.
Sources

Microsoft released Semantic Kernel python-1.46.0 on October 28, 2026, three weeks after 1.45.0, and the semantic-kernel package reached PyPI the same day. There are no new headline features. It is a hardening and bug-fix release, and several changes tighten what plugins and prompts are allowed to do.

Release facts

FactDetail
ProductSemantic Kernel, Python SDK
Version1.46.0 (GitHub tag python-1.46.0)
Release dateOctober 28, 2026 (GitHub release and PyPI upload)
Previous release1.45.0, October 6, 2026
LicenseMIT, per the GitHub API
Installpip install semantic-kernel==1.46.0

Security changes

  • Upload paths. SessionsPythonTool.upload_file rejects remote paths with dot segments such as ../; the .NET SDK got the same fix (PRs 14563, 14646).
  • HttpPlugin DNS pinning. Requests are pinned to the DNS addresses that passed validation, closing the gap between check and request (PR 14649).
  • No argument values in logs. Function argument values are no longer logged (PR 14574).
  • Prompt templates reject bytes in argument encoding, and the docs now spell out the prompt template trust boundary and tool invocation authorization (PRs 14669, 14648, 14647).

Connector fixes worth knowing

  • Anthropic: every stop reason is mapped instead of raising KeyError, and token usage is reported as CompletionUsage (PRs 14616, 14618).
  • Bedrock maps guardrail and context-window stop reasons; Gemini reports more finish reasons as content_filter; Mistral usage is a proper CompletionUsage object.
  • Ollama: streamed parallel tool calls keep their index (PR 14623).
  • Tool formatters for Anthropic, Bedrock, Google AI and Vertex AI honor include_in_function_choices (PR 14625).
  • Vector stores: Qdrant search works on qdrant-client 1.16 and later, Chroma and Weaviate stop doubling apostrophes in filters, multiple in-memory filters combine with AND, and PostgreSQL validates collection field types.
  • Kernel functions keep Pydantic field descriptions and Annotated metadata in schemas, and a None default is no longer reported as required.

Questions

Should I upgrade? Yes if you expose SessionsPythonTool or HttpPlugin to model-chosen input, or log at debug level in production. The release notes list no breaking change.

Is Semantic Kernel still maintained? It still ships: the release notes list 29 first-time contributors in this release alone. But the README now calls Microsoft Agent Framework its successor, so new agent features land there. See Microsoft Agent Framework 1.24.0 for the current release on that side.

What are the alternatives? Browse the frameworks category, including LangChain and AG2.

Alternatives

Semantic Kernel compared with its alternatives
ToolLicensePricing note
Semantic KernelThis launchSemantic Kernel is Microsoft's MIT-licensed SDK for AI agents and plugins in Python, .NET and Java, now succeeded by Microsoft Agent Framework.
MITSemantic Kernel is MIT-licensed and free to install from PyPI or NuGet (Java builds live in microsoft/semantic-kernel-java). You pay your model provider for tokens.
Microsoft Agent FrameworkMicrosoft Agent Framework is an MIT-licensed Python and .NET framework for AI agents and multi-agent workflows, and the successor to Semantic Kernel.
MITMicrosoft Agent Framework is MIT-licensed and free to install from PyPI or NuGet. You pay your model provider, or Azure if you host agents on Microsoft Foundry.
LangChainLangChain is an open-source agent engineering platform and Python framework for building agents.
MITNot stated

Questions

What is Semantic Kernel?

A model-agnostic SDK from Microsoft for building AI agents and multi-agent systems, with plugins from native functions, prompt templates, OpenAPI specs or MCP, plus vector store connectors. It supports Python 3.11 through 3.14, .NET 10+ and JDK 17+.

Is Semantic Kernel being replaced?

Its README says Semantic Kernel is now Microsoft Agent Framework and calls MAF the enterprise-ready successor, with a migration guide on Microsoft Learn. Semantic Kernel still gets releases: Python 1.47.0 shipped on November 5 and .NET 1.82.0 on November 7, 2026.

How do I install Semantic Kernel?

pip install semantic-kernel for Python, or dotnet add package Microsoft.SemanticKernel for .NET. Set OPENAI_API_KEY or your Azure OpenAI key first.

Should I start a new project on Semantic Kernel?

Microsoft's own pointer is to Microsoft Agent Framework for new work. Semantic Kernel makes sense if you already run it in production and want fixes while you plan a migration.

Filed on Agent Radar. Claims in this note are limited to the sources above.