惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

罗磊的独立博客
S
Secure Thoughts
C
Check Point Blog
D
DataBreaches.Net
V
V2EX
The GitHub Blog
The GitHub Blog
博客园 - Franky
博客园_首页
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Y
Y Combinator Blog
Last Week in AI
Last Week in AI
H
Help Net Security
WordPress大学
WordPress大学
酷 壳 – CoolShell
酷 壳 – CoolShell
宝玉的分享
宝玉的分享
A
About on SuperTechFans
月光博客
月光博客
Google DeepMind News
Google DeepMind News
小众软件
小众软件
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
F
Fortinet All Blogs
Blog — PlanetScale
Blog — PlanetScale
T
Tailwind CSS Blog
N
Netflix TechBlog - Medium
GbyAI
GbyAI
G
Google Developers Blog
P
Proofpoint News Feed
Recorded Future
Recorded Future
C
CXSECURITY Database RSS Feed - CXSecurity.com
MyScale Blog
MyScale Blog
T
Threatpost
V
Vulnerabilities – Threatpost
AI
AI
Cyberwarzone
Cyberwarzone
B
Blog RSS Feed
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
L
LINUX DO - 最新话题
L
LangChain Blog
Webroot Blog
Webroot Blog
AWS News Blog
AWS News Blog
云风的 BLOG
云风的 BLOG
W
WeLiveSecurity
Latest news
Latest news
U
Unit 42
C
CERT Recently Published Vulnerability Notes
Cisco Talos Blog
Cisco Talos Blog
www.infosecurity-magazine.com
www.infosecurity-magazine.com
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
S
Security @ Cisco Blogs
Forbes - Security
Forbes - Security

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor GitHub - GenAI-Gurus/awesome-eu-ai-act: Curated tools, official sources, OSS, templates, and guides for EU AI Act compliance. Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders How to Switch AI Chatbots and Why You Might Want To GitHub - MattMessinger1/agentic_refund_guardrail: Safe refund policy layer for AI agents — Python + TypeScript. Same behavior, shared tests. Adam/papers/emergent_values_whitepaper.md at master · strangeadvancedmarketing/Adam Ask HN: How do you stop playing 20 questions with your AI coding tools How far can automation and AI support psychotherapy? - @theU GitHub - stagas/rtdiff: realtime git diff gui and AI-assisted commits A Mac Studio for Local AI — 6 Months Later A History of the Early Years of AI at the University of Edinburgh Why AI Coding Tools Still Feel Stuck on Localhost MSN AI Datacenters Are Becoming Strategic Targets twitter.com Penn Researchers Use AI to Surface Unreported GLP-1 Side Effects in Reddit Posts Show HN: MoodSense AI (ML and FastAPI and Gradio, Deployed on Hugging Face) Moodsense Ai - a Hugging Face Space by aman179102 AI models are terrible at betting on soccer—especially xAI Grok GitHub - xialeistudio/echoic GitHub - HimashaHerath/github-dev-wrapped: AI-powered weekly GitHub activity reports deployed to GitHub Pages GitHub - alejandrobalderas/claude-code-from-source: Architecture, patterns & internals of Anthropic's AI coding agent — reverse-engineered from source maps AI and Tech brief: Ireland ascendant GitHub - Titovilal/context0: Context0 - Never Surrender Training for a Marathon with an AI Coach: What Worked and What Didn't Cyber Pulse: Agentic Intel - Apps on Google Play I Built an AI PR Reviewer That Catches Bugs by Not Looking for Bugs Gen Z workers are so fearful AI will take their job they’re intentionally sabotaging their company’s AI rollout | Fortune How AI Is Reimagining the Game of Golf–For Both Players and Courses GitHub - nattergabriel/reseed: A CLI tool for managing and distributing agent skills across projects Is SVG the final frontier? My AI workflow evolved from prompts to a near-autonomous workflow MLSharp Help - 3DGS Viewer & Generator I put my cognitive field based AI's runtime on GitHub Is Numble the first AI-proof game? A3: Kubernetes for autonomous AI agent fleets | Emergent Principles Deepali Vyas ("The Elite Recruiter") GitHub - msmarkgu/RelayFreeLLM: A restful API designed to route user prompts to various AI model providers. Unionized ProPublica staff are on strike over AI, layoffs, and wages Unleashing the Advantage of Quantum AI We're heading for an AI-fueled 'dementia crisis,' brain scientist warns The AI-Assisted Breach of Mexico's Government Infrastructure [pdf] GitHub - stef41/lmscan: 🔍 Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline. MSN GitHub - visionscaper/collabmem: Enabling long-term collaboration with Agentic AI - building up episodic and world model memory over time with in-context awareness We gave an AI a 3 year retail lease in SF and asked it to make a profit | Andon Labs AI Code is Hollowing Out Open Source, and Maintainers are Looking the Other Way What leaked "SteamGPT" files could mean for the PC gaming platform's use of AI AI is the boss at this retail store. What could go wrong? GitHub - Wuzu11517/agentic-proxy: Local proxy meant to help reduce With Drones, Geophysics and ArtificiaI Intelligence, Researchers Prepare to Do Battle Against Land Mines A Single Operator, Two AI Platforms, Nine Government Agencies: The Full Technical Report 在 Steam 上购买 FriedrichAI: Offline AI 立省 10% GitHub - inevolin/resume-cli: Hit Claude usage limits? Resume any AI coding session elsewhere. Switch tools at zero friction. GitHub - atripati/ark: AI Runtime Kernel — a context operating system for AI agents. Eliminates tool bloat, loads only what’s needed, and gives LLMs their reasoning space back. How to Build a Secure AI PR Reviewer with Claude, GitHub Actions, and JavaScript This Startup Wants You to Pay Up to Talk With AI Versions of Human Experts Intel Arc Pro B70 Brings 32GB VRAM to Local AI for $949 WordPress 7.0: The Good, the AI, and the Still Missing AI on the couch: Anthropic gives Claude 20 hours of psychiatry IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures AI Agents Know About Supabase. They Don't Always Use It Right. The history and future of AI at Google, with Sundar Pichai Inside an AI‑enabled device code phishing campaign How Meta Used AI to Map Tribal Knowledge in Large-Scale Data Pipelines AI for Systems: Using LLMs to Optimize Database Query Execution Forecasting the Economic Effects of AI Introducing Tinker: Play with AI, bring your ideas to life AI sheds light on an ancient gaming mystery People really hate AI but not as much as Iran—or Democrats | Fortune What is an AI Product Engineer? Phoebe Gates wants her $185 million AI startup to succeed with 'no ties to my privilege or my last name': 'I have a chip on my shoulder' | Fortune
I investigated Windows 11’s massive 5GB monthly .msu updates, and AI is only part of the problem
XzetaU8 · 2026-04-29 · via Hacker News - Newest: "AI"
Windows updates have grown drastically in size when compared to 2024
Windows updates have grown drastically in size when compared to 2024

As it runs in the background, you might not have noticed it already, but Windows 11 updates are getting huge. Monthly cumulative updates downloaded via Update Catalog are now regularly crossing 4GB, and in recent cases, even touching 5GB. The numbers get even more absurd when you look at what’s inside those updates. Our analysis showed a cumulative update expanding to nearly 9GB once extracted!

Note that this analysis is about the updates downloaded from the Microsoft Update Catalog, which is the original size of the update before it’s modified for your computer, depending on what hardware you have and which updates you’ve already installed. Windows Update applies hardware checks and pulls only the changed parts through Express updates and UUP. So, a 4GB catalog package can become a 1.5 to 2GB download at home.

Windows 11 24H2 cumulative update size over 19 monthly Patch Tuesdays, with the May 2025 jump highlighted

Either way, the actual update sizes have increased drastically, and the obvious explanation people jump to is AI. Windows is adding Copilot, on-device models, semantic search, and everything around it. So naturally, it must be AI bloating updates.

We found that it’s only part of the story.Windows update size increased from 300MB in 2024 to 5GB in 2026The real issue is how Windows updates are built and delivered. The size problem is about how new features are packaged, compressed, and distributed across millions of different systems.

I’ve always found it odd that an update that barely changes anything noticeable downloads gigabytes of data in the background. It’s also something to think of at a time when Storage is an expensive matter.

So, I did the research to find out why these Windows 11 updates are so big in the first place. And more importantly, can’t Microsoft just push updates only for the parts that change?

Why have Windows updates grown in size?

Microsoft doesn’t ship small, isolated patches anymore. We get the latest cumulative updates, also called LCUs.

So, the mandatory monthly security update (also called Patch Tuesday) includes everything. Not just the fixes from that month, but all previous fixes as well. If you install the latest update, your system is fully up to date, even if you skipped earlier ones.

Latest Windows 11 cumulative update goes above 5GB
Latest Windows 11 cumulative update goes above 5GB

It’s a clean model from a reliability standpoint. But the side effect is that updates can’t be small. They grow over time.

Even if your PC only needs a minor fix, the update package itself still carries the entire set of changes required to bring any system up to that version.

Microsoft tried to address this with Windows 11 24H2 through something called Checkpoint Cumulative Updates. Instead of building every update against the original RTM release, periodic “checkpoints” act as new baselines, and subsequent monthly updates only include changes since the last checkpoint, making them smaller and faster to install. Smart fix, on paper, at least.Each month measured against the September 2024 baseline

In practice, however, after a checkpoint was established in September 2024, updates were relatively small through April 2025. Then the May 2025 update nearly tripled in size, jumping to over 4GB in the catalog. A year later, there still hasn’t been a new checkpoint, which means the size savings Microsoft promised are quietly eroding.

Cumulative update till April 2025 was relatively smaller in size
Cumulative update till April 2025 was relatively smaller in size

Why Windows 11 May 2025 update tripled in size?

Windows 11 update for May 2025 was 3 times that of April 2025
Windows 11 update for May 2025 was more than 3 times that of April 2025

Looking more closely at the May 2025 cumulative update, the decompressed size jumped from around 6.5GB to nearly 9GB compared to April. What makes this unusual is that the compressed MSU file grew by roughly 3GB, while the decompressed file grew by only 2.5GB. Normally, compression makes the gap go the other way. The compressed file should always be smaller than the decompressed one, not larger in relative growth.

Compressed (MSU) vs. decompressed (extracted contents)
Compressed (MSU) vs. decompressed (extracted contents)

Windows updates contain tens of thousands of files, over 28,000 files in that single update. But none of the individual files stood out as unusually large. The top of the file size list was still the same msedge.dll from previous months.

Windows 11 24H2 May 2025 update extracted

The April 2025 cumulative was 1,287 MB. May 2025 was 4,369 MB. Our 7-Zip extraction of the May package shows dozens of MSIX files that were not present in April, including PSTokenizer, Text Recognition Session, PSOnyxRuntime, Query Processor Session, and Image Search Session, which are all components related to semantic search and on-device AI, using the Onyx runtime that Microsoft has also released for developers. The ~3 GB delta is attributable to those bundled MSIX payloads.

So yes, AI is definitely a factor for the update size increase. But there is more to it…

AI is in the package, but not on your PC

Windows Update uses applicability logic. Before downloading anything, it checks your hardware configuration and only pulls what your system actually needs. A fresh Windows 11 25H2 install on a VM with no updates installed, exactly the worst case for a large download, ended up downloading only around 1.7GB from a 4GB+ catalog package, because the Semantic Search MSIX components didn’t apply to that machine.

Looking at Delivery Optimization stats in Settings > Windows Update > Advanced options > Activity monitor confirms that the actual download size on real machines tends to be much smaller than what the Windows update catalog lists.

Activity Monitor for Windows updates

The Semantic Search components were initially built for Copilot+ PCs with Snapdragon NPUs. As support expanded to Intel and AMD systems, more variants got added to the package. Every expansion means more payloads in the bundle, even for machines that will never use those features.

Technically, this is solvable. AI models and Semantic Search components could be delivered through the Microsoft Store or as optional on-demand downloads, completely separate from the monthly cumulative update. Microsoft hasn’t done that yet, but they’re still shipping these components inside the main update pipeline.

Why can’t Microsoft just send exactly what changed in an update?

If Windows already knows what each system needs, why package everything together at all?

Differential updates already exist. Windows uses Express updates and the Unified Update Platform to pull only the changed portions of binaries rather than downloading entire files. A 3 to 4GB catalog entry may result in a 1.5 to 2GB download because of this. Microsoft is already minimizing what reaches your PC, but the inefficiency is in the package structure.

The May 2025 cumulative as it appears in the catalog vs. what reaches a typical home PC.
The May 2025 cumulative as it appears in the catalog vs. what reaches a typical home PC.

True modular updates are harder than they sound at the Windows scale (which is more than a billion users), with millions of hardware combinations, including different CPUs, GPUs, storage controllers, firmware versions, driver stacks, and legacy enterprise software. Every update has to be backward compatible, installable offline, and safe to deploy across thousands of machines without breaking anything. If a single dependency is missed or if it creates a component mismatch, the system can fail to boot.

There are also enterprise issues. WSUS downloads the full cumulative update every month. Configuration Manager distributes that same full package to every distribution point. Offline servicing tools inject complete MSU packages into system images. Every one of these scenarios needs the full package, even when an individual endpoint only uses a fraction of it. A home user might download 1.5GB, but an IT team managing thousands of machines is storing and distributing 4GB every month, per architecture.

Why Apple gets away with smaller updates

macOS incremental updates are typically in the 1 to 3GB range, with larger jumps reserved for major version releases. Apple controls the entire hardware stack, including CPUs, GPUs, drivers, and firmware. Older components can be dropped aggressively. Compatibility rules can be enforced strictly because the hardware variation is minimal compared to Windows.

Apple also doesn’t have a fixed monthly schedule. Updates ship when they’re ready, which allows more flexibility in how changes are bundled and what gets left for the next release.
Windows takes the opposite approach by prioritizing compatibility, predictability, and the ability to deploy across an enormous variety of hardware and software configurations.

What’s happening to Windows 11 updates in the enterprise infrastructure

For home users, the situation isn’t as bad as the catalog numbers suggest. Download sizes are much lower thanks to delta logic and applicability checks, which is why most people don’t notice the 3 to 4GB figure.

Enterprises don’t have that flexibility. Full packages get downloaded every month, stored across multiple distribution points, and pushed to endpoints regardless of what those endpoints need. Storage usage compounds quickly, especially when regular cleanup isn’t done. Distribution points push these packages across internal networks, and even with Delivery Optimization and peer-to-peer sharing, total data movement is significantly higher than it was two years ago. Compared to 2024, we are facing the worst memory crisis, and this isn’t a good sign for enterprises.

Enterprise cost has more than quadrupled

The yearly storage cost per architecture, per distribution point, has gone from roughly 11 GB (2024 cadence) to 52 GB (2026 cadence). For an org with even five distribution points, that is a quarter-terabyte of monthly update files sitting on disk per year, per architecture, without any cleanup.

Looking at 24H2 update sizes since launch, from around 200 to 500MB in mid 2024 to consistently 3 to 4GB by late 2025, the upward trend is clear and doesn’t look like it’s reversing anytime soon. Windows 10 went through a similar trajectory, just more slowly. Windows 11 is accelerating that pattern.

Win11 24H2 vs Win10 22H2 over the first 19 months of each release
Win11 24H2 vs Win10 22H2 over the first 19 months of each release

However, not all hope is lost, as Microsoft is taking efforts to reduce the download time for Windows updates, but we are not sure if it’s through size reduction.

How to check what your PC is downloading during an update

If you want to see the real download size rather than the catalog figure, go to Settings > Windows Update > Advanced options > Delivery Optimization > Activity monitor. It shows exactly how much data came from Microsoft, from local network peers, and from the cache.

For more details, Event Viewer includes Windows Update logs, and running Get-WindowsUpdateLog in PowerShell generates a readable log file.

Windows Update Log using PowerShell

The gap between what the catalog lists and what your machine downloads is usually the clearest way to show that the problem is the package design.

Windows updates are big because they are built for reliability

Windows updates are large because they’re built to work everywhere, on every configuration, for every enterprise scenario. Microsoft is already minimizing what reaches your device, but the package itself keeps growing. Until AI components ship separately and a new checkpoint resets the baseline, Windows updates aren’t going to get smaller.