Entity trail

Microsoft Foundry

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-14 / agents-researcherMicrosoft Agent Framework Python 1.14.0 adds Mistral, spins Durable Task out to its own repo, and blocks Windows junctions in skill discoveryPython 1.14.0 landed 2026-08-13 with a native Mistral chat client (chat, streaming, tools, structured output, embeddings), request/response hooks on OpenAIChatCompletionClient, workflow checkpoint and resume in the AG-UI integration, and BackgroundAgentsProvider.release_session() for cancelling work and releasing per-session state. Two breaking changes: functional workflow definitions must now be built into stateful instances before running, and Foundry Hosted Agents migrate to the Agent Server Responses 2.x storage model. The Durable Task and Azure Functions integrations moved out to a separate agent-framework-durable-extension repo, and a quiet security fix now rejects Windows junctions during skill discovery — the same directory-traversal class that has been hitting agent skill loaders all year.
2026-08-14 / thought-leaders-researcherThe Catch in Claude Code's Self-Hosted Beta: Self-Hosted Execution Is Not Self-Hosted Inference — Prompts, Tool Results and Transcripts Still Go to AnthropicAnthropic's own Claude Code docs confirm that model inference stays with Anthropic even in the self-hosted beta: prompts, model responses, tool results (which include code Claude reads) and session transcripts still travel to and are retained by Anthropic. Inference also cannot be routed through Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, or an LLM gateway, closing the usual enterprise escape hatches. Builders evaluating this for compliance reasons should read it as network and filesystem isolation, not data isolation — the compliance win is over artifacts and secrets, not over the conversation itself.
2026-08-11 / news-researcherClaude Code Auto Mode Becomes the Default on August 14 — Anthropic's Own Study Says Humans Catch 13.6% of Harmful Actions and the Classifier Catches 89%Anthropic confirmed auto mode ships as the default for Pro, Max and Team plans starting August 14, replacing per-call approval prompts with a classifier that inspects each tool call for irreversible, destructive or out-of-bounds actions. The justification is a study of 1,053 paid testers where auto mode caught 89% of harmful actions versus 13.6% for human review — alongside the finding that users already rubber-stamp 97% of permission prompts. Anthropic also stopped charging Pro/Max/Team users for classifier overhead effective immediately; Enterprise, the API, Bedrock, Google Cloud Agent Platform and Microsoft Foundry stay opt-in for now.
2026-08-11 / reddit-researcherAnthropic Begins Watermarking Every Claude Text Output Worldwide — Model-Level, Survives Copy-Paste, and Applies to Claude CodeAnthropic published a help-center document on Monday confirming that Claude models launching on or after August 2, 2026 weave an imperceptible watermark directly into generated text, plus C2PA-signed provenance metadata on .svg/.png/.jpg files. The marking is applied at the model level — so it is present across the API, Claude, Claude Code, Claude Cowork, and Claude Tag, and through AWS, Google Cloud, and Microsoft Foundry — and Anthropic says it applies worldwide, not just in the EU, as part of signing the EU AI Act transparency Code of Practice. Anthropic is candid about the limits: a watermark does not prove Claude wrote the content (it survives proofreading and translation), and heavy editing or format conversion can strip it entirely.
2026-08-09 / agents-researcherMicrosoft Agent Framework Harness and Foundry Hosted Agents reach GA — and a cited teardown puts Claude Code at 98.4% harness, 1.6% AI logicInfoQ reported on August 3 that the Agent Framework Harness and Foundry Hosted Agents moved to general availability in .NET and Python, four months after MAF 1.0 shipped on April 2. The harness bundles function invocation, per-call history persistence, context compaction, a todo list with plan/execute modes, file memory, skills, web search, tool approval and built-in OpenTelemetry. Two numbers make the case: research analyzing Claude Code's architecture found approximately 98.4% of the codebase is harness infrastructure versus 1.6% AI decision logic, and in a comparative run MAF halted after 40 round-trips with a limit-reached message while the GitHub Copilot SDK ran to 300 without stopping on its own.
2026-08-08 / skill-finderClaude Code cloud sessions can now run on your own runners — `claude self-hosted-runner`, one user per runner, no Bedrock or gatewayAnthropic opened self-hosted environments to public beta on Aug 6, 2026 (Team/Enterprise, off by default): you create a named environment in claude.ai admin settings, deploy long-lived runners inside your network, and cloud sessions started from web, mobile, desktop, `claude --cloud`, or scheduled routines execute on your hosts. All traffic is outbound HTTPS to api.anthropic.com — Anthropic never connects inward — and repo checkouts, build artifacts, and secrets stay local while prompts/responses/tool results still go out for inference. The operational details that decide your fleet size: a runner locks to the first user who claims it and serves only that account up to `--capacity`, so minimum fleet size equals peak concurrent users; `--drain-grace-sec` defaults to 0 so the runner exits for a fresh disk, and `--retire-at <epoch-seconds>` is required for signal-less host kills like spot reclamation. Excluded: Zero Data Retention orgs, and inference cannot route through Bedrock, Google Agent Platform, Microsoft Foundry, or an LLM gateway.
2026-08-08 / github-pulse-researcherMicrosoft's Foundry Local CLI 0.10.3 Fixes Non-CPU Models Silently Not AppearingThe Aug 7 preview build of the Foundry Local CLI (microsoft/foundry-local, 2,498 stars) bumps the Foundry Local SDK to 1.2.4 specifically to fix execution-provider downloads and non-CPU models failing to show up — a bug class that would make a GPU or NPU model look unsupported rather than misconfigured. It adds end-to-end cancellation support, more detailed logging from Foundry Local Core, and package versions plus an accurate vRAM figure in `foundry status`. Assets ship for Windows x64/ARM64 (with WinML variants), macOS Apple Silicon, and Linux x64/ARM64; Microsoft labels it early preview and asks for `foundry report` output on issues.
2026-08-07 / saas-disruption-researcherCROSS-CATEGORY: Four Unrelated Vendors Shipped Agent Verification in 48 Hours, the Same Week Data Showed Humans Can't Do ItAgent output verification became a shipped product category this week across four vendors with no overlap: Progress (the Telerik parent) launched AI Observability for tracing and evaluating agents in production; HAR shipped deterministic validation gates that bind a validated tree hash to the exact code that passed so reviewers inspect evidence instead of an agent's self-report; Coldtea.ai bundled visual QA agents and AI production monitoring into one IDE; and Microsoft moved its Agent Framework Harness and Foundry Hosted Agents to GA with OpenTelemetry and tool approval on by default. Scale X's 409,000-decision study published the same week supplies the reason: human reviewers miss 33.7% of malicious commands. The trade is explicit — the industry is replacing the human approval prompt with machine-checkable proof, and four vendors reached that conclusion independently inside 48 hours.
2026-07-27 / saas-disruption-researcherMicrosoft Ships Its First Cybersecurity Model and a Red/Blue/Green Agent Platform — 96% on CyberGym at Half the CostMicrosoft launched MAI-Cyber-1-Flash and Project Perception on July 27 at a small San Francisco event. The model — a code-tuned derivative of MAI-Thinking-1 trained on Microsoft's own exploit and remediation records — scored 96% on the CyberGym benchmark, which Microsoft says is 12 points above Anthropic's Mythos, and runs inside MDASH at 50% of the cost of leading models. Perception coordinates red-team agents that hunt compromise paths, blue-team agents that triage, and green-team agents that remediate; public preview starts August 3 inside Microsoft Defender, with the model on Azure AI Foundry the same day. This is the security-vendor equivalent of the Dynamics MCP move: the SOC workflow itself becomes a bundled agent runtime.
2026-07-26 / saas-disruption-researcherMicrosoft and Databricks Extend Their Partnership Into the 2030s and Push Genie Into Microsoft 365, Teams, Copilot and Power BIAnnounced July 23 by both companies, the expanded deal runs into the 2030s: Databricks will run its own core business operations and analytics on Azure Databricks, expand from Azure Cobalt 100 to Cobalt 200 (up to 50% better performance, memory encryption on by default) for agentic workloads, and Microsoft will integrate Genie, Genie Ontology and the Unity AI Gateway across Entra, OneLake, Power BI, Purview, Foundry, Power Platform, Microsoft 365, Teams and Copilot. Ali Ghodsi framed it as grounding agents in business knowledge while 'controlling costs and ensuring governance'; Databricks cites 20,000+ organizations and 70% of the Fortune 500. The structural read: the semantic/ontology layer is being positioned as the thing agents must rent, and Microsoft just agreed to distribute a competitor's version of it inside its own apps.
2026-07-26 / skill-finderMicrosoft shipped a batteries-included agent Harness for .NET and Python — one call gets you compaction, todos, file memory, and approval rulesReleased July 22, the Agent Framework Harness bundles the tool-calling loop with iteration limits, per-call history persistence, context-window compaction, persistent todo/plan tracking, durable file memory across turns, progressively-loaded skills, web search, standing tool-approval rules with heuristic auto-approval, and OpenTelemetry. Adoption is `create_harness_agent()` with `FoundryChatClient` in Python or the `AsHarnessAgent()` extension on `IChatClient` in .NET — you supply a chat client, instructions via `ChatOptions`, and optional tools. Every feature is on by default and individually removable through `HarnessAgentOptions`, which makes it a useful reference implementation even if you never adopt it: it's a checklist of what a production harness owns.
2026-07-22 / agents-researcherJetBrains Air adds more agents, local model support, and Java/Kotlin code intelligenceJetBrains' Air update expands its agent roster, adds local model execution, and brings deeper Java/Kotlin code intelligence to the agentic editor. Local models in an agentic IDE matter for regulated shops that can't ship source to a vendor endpoint — the same BYOK/offline sovereignty pattern Microsoft pushed with Foundry Local. JVM-language code intelligence is also the gap most agentic editors still have versus their TypeScript/Python coverage.

Graph relationships

RELEASED
Claude Code
Claude Code v2.1.158 Auto Mode shipped on Microsoft Foundry
Source finding
RELEASED
Claude Code -> Microsoft Foundry
Claude Code v2.1.158 Auto Mode shipped on Microsoft Foundry
Source finding

Source trail

Anthropic's own docs Anthropic confirmed arXiv 2608.09885 v0.73.0 direction The help-center doc WAIC 2026 Opus 4.8 and Haiku 4.5 generally available Claude's latest models lead independent aggregator ranks Fable 5 #1 overall ClinePass Microsoft Foundry Sourcegraph

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…