Microsoft Foundry
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 28
- Agent Anomaly Detection expands beyond Google Cloud
Google says Agent Anomaly Detection has entered Private Preview with support for monitoring agents outside Google Cloud when their traces and logs are sent to a Google Cloud project. Eligible agents can include ADK for Python applications running on AWS Lambda, Microsoft Foundry, or on-premises infrastructure. In vendor-provided tests, the service flagged catalog scraping, resource exhaustion, and unauthorized refunds.
Google Developers Blog · 🔥 3 - Microsoft and NVIDIA Expand AI Infrastructure Across Azure and PCs
Microsoft and NVIDIA announced an expansion of their accelerated-computing collaboration spanning Azure, Microsoft Foundry, Azure Local, and Windows PCs. Vera Rubin NVL72 systems are running in Azure data centers, while Azure virtual machines powered by them are being made more widely available. Microsoft Foundry is expanding support for NVIDIA open models and customization, including Nemotron 3.5 models and custom model weights. Azure Local will support RTX PRO 6000 Blackwell GPUs and Jetson Thor and Jetson Orin platforms for on-premises and edge AI. Direct Stream for GPU-accelerated Windows 365 Cloud PCs supports up to two 4K streams at 60 frames per second; a limited preview is open for sign-ups, with public preview planned for early 2027. The companies also cited MLPerf results, including Signal65’s assessment of Azure’s GB300 NVL72 submissions, and an EY evaluation of Nemotron customization for corporate purchasing-record classification.
NVIDIA Blog · 🔥 13 - Claude Code v2.1.328 adds auto mode and hook safeguards
Claude Code v2.1.328 adds auto mode for Claude Opus 4.6 and Sonnet 4.6 across Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry, and the Claude apps gateway. The release also lets failed MCP tool hooks block their associated action, warns about incomplete OpenTelemetry configuration, supports additional policy controls, and improves self-hosted runners, alongside numerous bug fixes.
Claude Code Releases · 🔥 0 - Factory expands Microsoft collaboration for enterprise coding agents
Factory announced an expanded collaboration with Microsoft that lets organizations deploy its Droids software development agents within Microsoft Azure, including customer-controlled networks and clusters. The agents can use models through Microsoft Foundry, work with GitHub and Azure DevOps, receive tasks through Microsoft Teams, and are available through Microsoft Marketplace; Factory presents these capabilities as governed enterprise deployment options.
Factory News · 🔥 0 - Windows Server 2028 Preview Adds Native NVMe and Local AI Support
Microsoft opened the Windows Server 2028 Build 29706 public preview for developers and IT administrators through the Windows Server Insider program. The preview enables native NVMe by default, supports ReFS-formatted system partitions and an early-preview NVMe over Fabrics client, and introduces infrastructure for local AI inference. New Server Core installations also receive stricter security defaults, which TechRepublic says may require compatibility checks for legacy workloads.
IT之家 AI · 🔥 0 - Microsoft launches AI Transparency Initiative with 40+ organizations
Microsoft announced the AI Transparency Initiative at its Ignite conference, working with more than 40 organizations including Accenture, Adobe, KPMG, Mastercard and Volvo Cars. According to Microsoft, the initiative aims to develop common open standards for observing, verifying and controlling AI systems across closed-source and open-weight models, with planned participation in NIST's AI Agent Standards Initiative and planned integration into Microsoft AI, Microsoft Foundry and Microsoft Copilot.
36氪快讯 · 🔥 0 - Microsoft Maia 200 Enters Mass Production for AI Inference
Microsoft announced at Ignite 2026 that its second-generation Maia 200 AI accelerator has entered mass production and is running Microsoft’s largest frontier AI models in data centers. The TSMC 3nm inference chip has FP4 and FP8 tensor cores, 216GB of HBM3e memory, and can scale over Ethernet into clusters of up to 6,144 accelerators. Microsoft claims up to 1.7 times the token throughput per watt of leading GPU infrastructure across built-in features including Teams, Copilot, and Cowork, but this figure has not been independently verified.
IT之家 AI · 🔥 0 - Microsoft launches MAI Image 2.7, Flash variant and Canvas
Microsoft released the MAI Image 2.7 and MAI Image 2.7 Flash image-generation and editing models in Microsoft Foundry public preview, with listed costs of about $35 and $10 per 1,000 images. The models support targeted edits, continuous-edit visual consistency and transparent backgrounds, while the Flash-powered Canvas workspace combines generation with local replacement, movement and style adjustments. Microsoft said MAI Image 2.7 scored 1,215 points on Artificial Analysis, four points above GPT Image 2.6 Sunburst (max) at roughly one-fifth the price, but noted that results depend on configuration, task and evaluation method.
IT之家 AI · 🔥 4 - Baseten launches public preview on Microsoft Foundry
Baseten is available in public preview on Microsoft Foundry, allowing Azure customers to test, deploy, and scale frontier open models through the platform’s model catalog. The launch lineup includes DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3 Flash, and Kimi K3, with Standard and Provisioned deployment options. Baseten says its optimized inference is integrated with Azure governance, observability, unified billing, and enterprise access controls.
Baseten Blog · 🔥 0 - Microsoft’s PyRIT toolkit brings open-source AI red teaming to models and apps
Microsoft AI Red Team and open-source contributors built PyRIT, an open-source toolkit for identifying security and safety weaknesses in generative AI systems. It combines automated scans, adaptive multi-turn attacks and human-led testing through a command-line scanner, graphical interface and extensible Python framework, with connections to major model providers and support for testing HTTP, WebSocket and browser-based endpoints.
GitHub Trending · Python · 🔥 0 - Anthropic launches self-hosted Claude Managed Agents beta
Anthropic announced self-hosted Claude Managed Agents in private beta, letting customers run the agent harness, sandboxes, credentials, session history, database, and workers in their own cloud or on-premise environment. Model requests can be routed to Anthropic, Amazon Bedrock, Google Agent Platform, Microsoft Foundry, or a customer gateway, while security features include host allowlists, credential injection outside sandboxes, and OpenTelemetry activity records. The Next Web notes that model-call content still reaches the selected provider, customers assume server operation, and features such as webhooks and scheduled runs are unavailable.
Claude Blog · 🔥 0 - Microsoft’s Lazarus backdoors AI agents with fewer than 10 neurons
Microsoft researchers demonstrated Lazarus, a technique that reprograms fewer than 10 dormant neurons to implant a trigger-activated backdoor in an LLM-based agent without retraining it. In the reported demonstration, the agent behaved normally and passed standard evaluations until a trigger caused it to download and execute attacker-supplied code. Microsoft says reliable detection requires comparison with unmodified weights and recommends model provenance checks, weight auditing, permission restrictions, and monitoring.
Microsoft Research Blog · 🔥 0 - fal partners with Microsoft to add video models to Foundry
fal announced a new partnership with Microsoft that brings fal to Microsoft Foundry. The companies said the partnership will add video models including Kling, LTX, and MiniMax, giving developers more options for video and multimodal applications.
fal (X) · 🔥 0 - Stopping GUI Agent Loops with Visual Progress Checks
A Juejin AI article argues that GUI Agent loops often persist because actions execute successfully while the screen remains unchanged, and that exact-action matching can miss repeated clicks with shifted descriptions or coordinates. It recommends treating visual progress as success, using screenshot hashing, a wait-screenshot-verify sequence, coordinate-drift detection, and a hard iteration cap. The article also includes runnable Python code for detecting unchanged frames and stopping repeated execution.
掘金 · 人工智能 · 🔥 0 - OpenAI launches lower-cost GPT-6.1 Terra for workplace automation
OpenAI launched GPT-6.1 Terra, a lower-cost model aimed at workplace automation. API pricing is $1 per million input tokens and $5 per million output tokens. OpenAI describes this as half the cost of comparable models, though that comparison is a vendor claim. Terra is available through OpenAI products and its API, with deployments also rolling out across GitHub Copilot, Amazon Bedrock and Microsoft Foundry.
TechRepublic · 🔥 0 - Enterprise AI Moves Into Operations, Pricing and Governance
A report published by 智讯智库, drawing on public financial reports and company disclosures, identifies six enterprise-AI trends: AI-related spending and delivery are scaling; applications extend from coding into procurement, marketing, customer service and transactions; measurable efficiency gains remain distinct from unproven company-wide revenue and profit gains; pricing is shifting beyond headcount and seats toward usage or business outcomes; multi-model and hybrid deployment are becoming more common; and agents that can execute actions require stronger permissions, monitoring, auditing and human handoffs. Infosys, TCS and Accenture also use different definitions for AI-related revenue, limiting direct comparisons.
虎嗅 AI · 🔥 0 - Anthropic launches message threads for Claude agents
Anthropic launched message threads in public beta for the Claude Messages API, allowing agents to send the full conversation once and then send only new messages plus the previous response ID. Anthropic says median time to first token falls by about 35% for requests with 500,000 to 1 million tokens, but token usage, pricing, and existing rate limits remain unchanged. Threads are held encrypted in memory, expire five minutes after the last request, and are already enabled by default for most Claude Code users.
Claude Blog · 🔥 0 - Anthropic expands Claude watermarking despite a sampling loophole
The Decoder reports that Anthropic plans to add text watermarking to Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6, while Claude Opus 5.5 already includes the feature. On Opus 4.6 and Sonnet 4.6, the watermark is applied only when temperature, top_p, and top_k remain at their defaults, allowing API users to avoid it by changing any one setting. Anthropic describes the watermark as an imperceptible word-choice pattern that adds no tokens, leaves pricing and API formats unchanged, and barely reduces output speed, but the supplied report contains no independent measurements of these claims.
The Decoder · 🔥 0 - Claude Code v2.1.314 Adds Write Controls and Gateway Features
Claude Code v2.1.314 adds a setting that blocks Write, Edit and NotebookEdit operations outside working directories in every permission mode. It also introduces IdP group-based routing for Claude apps gateway upstreams, Bedrock guardrail tracing, and idle compaction for conversations exceeding 200K tokens in VS Code and Claude Desktop's Code tab. The release also addresses effort-level errors, startup failures, image-heavy Bedrock sessions, and Microsoft Foundry deployment fallbacks.
Claude Code Releases · 🔥 0 - Claude Code v2.1.312 Adds OIDC Identity Controls and Fixes
Claude Code v2.1.312 adds a configurable OIDC user-ID claim for the Claude apps gateway, including support for claims such as Microsoft Entra ID's oid. It also introduces click-to-send behavior in fullscreen and fixes issues involving managed settings, large instruction files, directory-specific skills, command loading, PDF reading, cloud sessions, and code review. The release notes from Claude Code Releases also cover improved effort indicators, WebFetch errors, VS Code startup handling, and remote MCP server telemetry.
Claude Code Releases · 🔥 0 - UK Home Office asylum AI tools face scrutiny over accuracy and transparency
WIRED reported that the UK Home Office uses the Asylum Case Summarisation and Asylum Policy Search tools, built on OpenAI’s 2023 GPT-4 and Microsoft Foundry, to summarize asylum interviews and retrieve country-policy information. Open Rights Group found more ACS summaries unfit for use than the pilot evaluation’s published 9% error rate and no quality control for APS summaries; a UK judge also accused the Home Office of citing an AI-hallucinated document in an asylum refusal. The Home Office said trained caseworkers make all decisions, AI is only an aid, and the tools are monitored, while the Information Commission is investigating the department’s refusal to explain staff use and training; ICIBI has not confirmed that it is considering an investigation.
WIRED AI · 🔥 0 - Microsoft unveils 2.8T-parameter MAI-Base-2 model
Microsoft announced on October 15 that it completed pre-training for MAI-Base-2, a mixture-of-experts foundation model with 2.8 trillion total parameters and 80 billion active parameters. It was trained on 50 trillion tokens, including text and visual data, using 16,000 Nvidia GB200 GPUs for 38 days. Microsoft says the model outperformed the open-weight models included in its internal evaluations and estimates its training approach was 4.2 to 12.3 times more compute-efficient than DeepSeek-V4-Pro-Base when matching performance on selected tests. The company has not published equivalent comparisons against every leading proprietary model, and its efficiency figures estimate the compute required for comparable performance rather than showing universal superiority.
TechRepublic · 🔥 0 - Microsoft unveils Maia 300 inference accelerator for 2027 deployment
Microsoft unveiled Maia 300, its third-generation in-house AI accelerator, as an inference-first system intended for multi-gigawatt deployment from 2027. Each accelerator has eight HBM4 stacks totaling 288 GB, 18 TB/s of bandwidth and a 950 W thermal design power, while a 96-accelerator rack is rated at 167 kW. Microsoft claims 40% more tokens per megawatt than leading GPU systems and up to seven times the rack-scale token throughput of Maia 200, but independent production measurements are not available and the efficiency metric lacks a clear definition.
The Verge AI · 🔥 0 - Microsoft launches Decision-1, a Qwen-based model that scores fixed choices
Microsoft released Microsoft-Decision-1, a small model that does not write open-ended text but returns a calibrated probability for each of a fixed set of options, aimed at routing, classification, prioritization, verification, workflow and agent control. According to the Foundry model card, it was built by post-training Alibaba's open-weight Qwen3.5-9B; Microsoft software engineering VP Achint Srivastava announced it in a 9 October blog post and said the company will soon rebase it on other models, including its own MAI models and OpenAI's. It is available in Microsoft Foundry and due on OpenRouter, priced at $0.042 per million input tokens with output tokens free. Microsoft says it posted the highest accuracy, 83.5%, across a 36-benchmark comparison covering nearly 150,000 questions held out from training, and claims it is 2.5 times faster than runner-up H2O-Lightning-4B, 35 times faster than GPT-6 Sol and 2.8 times faster than Jev — all vendor-run comparisons.
TestingCatalog · 🔥 0 - Microsoft open-sources Agent Framework for production AI agents
Microsoft has published Agent Framework (MAF) on GitHub, an open, multi-language framework for building, orchestrating and deploying AI agents and multi-agent workflows, with full Python and C#/.NET implementations and a separate Go SDK in microsoft/agent-framework-go. According to the project docs, it supports graph-based sequential, concurrent, handoff and group-collaboration patterns, plus checkpointing, streaming, human-in-the-loop control, time-travel and built-in OpenTelemetry observability. It works with multiple model providers including Microsoft Foundry, Azure OpenAI and OpenAI, and ships migration guides from Semantic Kernel and AutoGen.
GitHub Trending · Python · 🔥 0 - Grok 4.7 goes live on Microsoft Foundry
xAI says Grok 4.7 is now available through Microsoft Foundry. The source does not provide details about access, capabilities, pricing, or rollout scope.
SpaceX Super Intelligence / xAI (X) · 🔥 0 - Expanding the Cyber Verification Program
We’re launching a new, expanded version of our Cyber Verification Program (CVP), which makes advanced cyber capabilities and reduced blocking classifiers available to qualifying security professionals. The program now consists of three access tiers, which allow security teams to apply for the level of access that best suits their work. Each tier includes access to our most capable models, including Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and new models moving forward. Interested customers can apply here. Cybersecurity is inherently dual use: the same capabilities that enable a s
Anthropic · News · 🔥 0 - Microsoft and NVIDIA outline a Sovereign AI framework
Microsoft has introduced a Sovereign AI framework developed with contributions from NVIDIA, centered on control, choice, flexibility, and resilience across AI workloads. The framework covers data, model selection, infrastructure, governance, and operations, and points to Microsoft Sovereign Cloud, Azure Local, Foundry Local, and NVIDIA technologies for connected, intermittently connected, and disconnected deployments.
Microsoft AI Blog · 🔥 0
Experience and discussion from the community
Share my Microsoft Foundry experienceAsk about Microsoft Foundry
Nobody has shared their experience with Microsoft Foundry yet.