Loading story…
Microsoft unveiled Project Zenith, a developer-focused Windows 11 configuration for high-memory PCs that can run AI models exceeding 30 billion parameters locally. The initiative targets systems with at least 64 GB of unified memory and more than 250 GB/s memory bandwidth; initial devices will use AMD Ryzen AI Halo, with additional OEM and silicon partners expected.
Zenith combines Windows performance tuning and developer-oriented defaults with integrated WSL containers and Microsoft Execution Containers. Microsoft said the configuration provides containment, identity controls, and enterprise management for agentic-development workloads, enabling routine AI-assisted development to move from metered cloud inference to on-device models while retaining cloud services for more demanding tasks.
Track how attackers are adapting to this technology.
3 events from the most recent confirmed update back to the earliest known activity.
At IFA 2026, Microsoft presented its "unmetered intelligence" vision for Windows, emphasizing local execution of AI workloads on CPUs, GPUs, and NPUs alongside cloud services. The company also described plans for Windows AI-agent tooling, isolation containers, and future auditing, logging, and policy controls.
Microsoft said it developed Microsoft Execution Containers with NVIDIA and silicon partners to isolate, apply policy to, and log autonomous Windows agents. NVIDIA plans to integrate its OpenShell agent framework with the containers; Hermes Agent and OpenClaw were identified as OpenShell integrations.
Microsoft introduced Project Zenith, a developer-focused Windows 11 configuration for high-memory PCs intended to run AI models exceeding 30 billion parameters locally. The configuration targets systems with at least 64 GB of unified memory and more than 250 GB/s memory bandwidth, initially using AMD Ryzen AI Halo, and includes WSL containers and Microsoft Execution Containers for managed agentic-development workloads.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
6 references tracked. Mallory keeps watching after this page renders.
windowslatest.com
Open sourcehelpnetsecurity.com
Open sourcewindowslatest.com
Open sourcetomshardware.com
Open sourcecryptika.com
Open sourcecybersecuritynews.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Huawei is advancing its domestic semiconductor ecosystem with the Ascend AI chip family, positioning its CloudMatrix 384 system as a direct alternative to Nvidia's GB200 platform. The Ascend 910C, a dual-chiplet accelerator with stacked HBM2E memory and DaVinci NPU architecture, is at the core of this effort, delivering up to 780 TFLOPS of dense BF16 compute. While the chip lags behind Nvidia's Hopper and Blackwell architectures in peak throughput and power efficiency, Huawei is compensating by scaling up deployments and building a robust supplier network, with over 60 semiconductor companies backed by its Hubble investment arm. This initiative is part of China's broader push for semiconductor independence, with local partners developing design toolchains and software ecosystems to reduce reliance on U.S. vendors. Meanwhile, AMD is preparing for the next generation of AI and HPC workloads by upstreaming Linux kernel patches for its upcoming Instinct MI400-series accelerators, likely based on the CDNA5 architecture. These patches indicate hardware enablement for new GPU IP blocks, including platform security and memory controllers, and suggest that AMD is moving forward with products intended to compete in the high-performance AI accelerator market. The developments from both Huawei and AMD highlight the intensifying global competition in AI hardware, with each company pursuing different strategies to address performance, efficiency, and supply chain independence.
Jun 29, 2026
OpenAI and AI chipmaker **Cerebras** announced a three-year strategic alliance under which OpenAI plans to deploy up to **750MW** of Cerebras *Wafer-Scale Systems* to build a large high-speed AI inference platform; reporting cites the deal value as **over $10B** and claims up to **15x faster** large language model response times versus GPU-based alternatives. Separately, China-based **Zhipu AI** reported it trained an image-generation model using only **Huawei Ascend Atlas 800T A2** hardware and the *MindSpore* framework, positioning the result as evidence that advanced model training can proceed without access to Western GPUs amid US restrictions that have limited Chinese firms’ access to **NVIDIA H100/A100**-class accelerators. In parallel, security testing of popular “vibe coding”/AI coding assistants found they can repeatedly generate **unsafe code** from common prompts, including issues rated **critical** in severity. A security startup (Tenzai) evaluated tools including **Claude Code**, **OpenAI Codex**, **Cursor**, **Replit**, and **Devin** by using predefined prompts to build test applications, concluding that while some tools avoid more rule-based defects, they struggle in context-dependent security boundaries—raising risk that AI-assisted development could introduce exploitable vulnerabilities into production code if not paired with strong review and secure-by-design controls.
Jun 29, 2026
Perplexity has released **Portable Computer**, a Windows version of its Computer agent that runs locally on compatible Nvidia GeForce RTX and RTX PRO GPUs with at least 24 GB of VRAM. Available to Perplexity Pro and Max subscribers, the agent supports multistep task planning, browser use, tool calling, connectors, scheduled recurring tasks, local MCP servers for desktop applications, and one-click downloads of local models including PPLX 27B and Qwen 3.8 27B. Although inference can run on-device and does not consume Computer credits, the deployment is not fully air-gapped: Portable Computer can connect to local and external resources such as Microsoft and Google productivity services, Slack, and GitHub. It may also ask users to authorize sending work to Perplexity cloud models when additional reasoning capacity is required, creating a hybrid local/cloud data-processing path that organizations should assess before connecting enterprise resources.
Sep 15, 2026