Microsoft, Apple and Amazon AI

Oct 9

Apple Machine Learning Research published Normalizing Trajectory Models, a way to generate images in four steps while keeping an exact likelihood. This post explains the idea in plain language, the reported results, and what the paper does not claim.

Oct 8

Zenity Labs published AgentCorruption, a chain of flaws in Amazon Bedrock AgentCore that let a single prompt to one public agent become control over every agent in the same AWS account and region. AWS has since patched parts of it. This post explains the chain, the timeline, what is still your job, and a hardening checklist.

Oct 8

At its October 7, 2026 Windows event Microsoft announced a local build of MAI-Code-1.1 Flash, a 137B-parameter coding model with a 256K context window, and said GitHub Copilot will hand work to local models. Here are the numbers, the hardware you need, and the claims to verify before you plan around them.

Oct 8

On October 7, 2026 Microsoft framed Windows as a platform for hybrid intelligence: agents that run on the device when that fits and in the cloud when needed, inside containers that limit what they can touch. Here is what was announced, what is still future tense, and what developers should do now.

Oct 7

Microsoft Research released Agent Lightning v1.0, an MIT-licensed framework for reinforcement learning on agents that keeps the deployment harness intact. This post explains the idea, the numbers, the hard engineering problems the paper names, and what builders can take from it.

Oct 7

Microsoft has put prices on its Nvidia RTX Spark hardware: the Surface RTX Spark Dev Box is $5,999 and ships in November, and the Surface Laptop Ultra starts at $2,599 from October 16. Here is what you get, how the price compares with the rest of the local-AI box market, and who should wait.

Oct 5

RemoveMacAI is an open-source CLI that turns off Apple Intelligence on macOS 27, deletes its downloaded models and stops macOS from fetching them again. A 367-point Hacker News thread argued about storage, consent, soldered SSDs and curl-pipe-bash. Here is what it changes, how it works, the risks, and what you can do without it.

Oct 4

Microsoft researchers released ThinkingBox, an open sandbox and benchmark that checks what an agent actually changed in a backend instead of what it said. Across 121,680 trials, 79,853 failed attempts, and two-thirds of those ended with no tool error. Here is what the numbers show and how to apply the idea to your own agents.

Oct 3

After an Inc columnist said Meta's Muse read his Mac Messages without Full Disk Access, Apple said it will add controls so only very explicit user action can grant an app that level of access. Meta says its opt-in gates make that impossible. Here is what is established and what to check on your own Mac.

Oct 2

On October 1, 2026, Artificial Analysis ranked Microsoft AI's MAI-Transcribe-2-Streaming #1 of 38 on AA-WER Streaming: 2.5% word error at 0.13 seconds after end of speech. Mustafa Suleyman called it the most accurate real-time transcription model and asked builders to come build agents. The leaderboard backs the accuracy claim. It does not back the "60% cheaper than ElevenLabs" line.

Oct 1

Effective October 7, 2026 Amazon EC2 Capacity Blocks for ML charge more per reserved NVIDIA accelerator. On-Demand and Savings Plans are unchanged, the rate is locked when you purchase, and Trainium blocks are not in this list. If you already planned a GPU block, the calendar date is the budget event.

Sep 24

At Accelerate, Amazon upgraded Seller Assistant with persistent memory, 24/7 workflows, a visual canvas and a plugin that lets sellers manage inventory, pricing and listings from Claude or Amazon Quick. Headlines say agents now run operations for 90% of sellers. The 90% figures actually mean something else. Here is the accurate version.

Sep 24

The Information reports Microsoft will discount Copilot by 30% for customers with 1,000 to 10,000 seats and 50% at 10,000 or more, starting around October, as it merges chat, Cowork, Autopilot and Code into a super app and moves to seat-plus-usage billing. Here is what is confirmed, what is reported, and how to negotiate.

Sep 23

Running a model with a trillion parameters on consumer-adjacent hardware is either a genuinely remarkable feat of quantization and unified-memory engineering, or a demo built on assumptions (heavy quantization, narrow benchmark, cherry-picked setup) that don't survive contact with real workloads. Here's what the claim requires to be true, and what's missing to fully evaluate it.

Sep 23

John Ternus, now Apple's CEO, posted "a huge leap in performance, especially for AI" about the Mac mini and Mac Studio on September 22 — nearly a month after Apple actually unveiled the M6 Mac mini and M5 Ultra Mac Studio on August 25. The confusing part holds either way: the base Mac mini ships on the newer M6 chip while the flagship Mac Studio runs the previous-generation M5 Ultra. Here's what that means for local AI, and why Ternus revived the pitch a month later.

Sep 17

Apple is reportedly developing its own AI server hardware built around NVIDIA's NVLink Fusion interconnect technology — described as Apple's first return to building server-class hardware since 2011, a notable strategic shift for a company that has historically relied on external cloud providers for AI training infrastructure.

Sep 14

A code sleuth found private frameworks in iOS 27 and macOS Golden Gate showing Apple built Siri to let Claude appear as a Siri extension, or even replace Siri's own server-side model entirely with something like GPT-5.6. Neither is user-facing yet, but the plumbing is real — here's what Model Delegation and Model Manager Services actually do.

Sep 14

Mustafa Suleyman posted Microsoft AI's "Code of Conduct for Humanist AI" on September 14, 2026, open for public comment for six weeks. It rejects model welfare and AI rights outright, bans "neuralese" reasoning humans can't read, and requires every MAI model to be interruptible, correctable, and shut-down-able before it ships. Here's what each of the 10 points actually commits Microsoft to.

Sep 14

Microsoft CEO Satya Nadella posted a principle on September 14, 2026 — any pursuit of superintelligence must keep AI helpful and under human control, with benefits diffused broadly rather than held by a handful of labs. Backing it, Microsoft is publishing a "Code of Conduct" for its first-party MAI models for public consultation, alongside enterprise control over learning loops. Here's what he actually committed to versus what's still just a principle.

Sep 13

Microsoft announced on September 12, 2026 that Grok models are now available as a preview option inside Copilot for Word, Excel, and PowerPoint — rolled out through Microsoft's Frontier Program, off by default, and requiring a separate admin setting. It's the clearest sign yet that Microsoft is treating Copilot as a multi-model surface rather than a Microsoft/OpenAI-only product.

Sep 11

Sep 10

On September 9, 2026, John Ternus delivered his first keynote as Apple CEO — eight days after Tim Cook handed over day-to-day leadership. The lineup: iPhone Duo, Apple''s first foldable at $1,999; iPhone 18 Pro and Pro Max from $1,199; Apple Watch Series 12 and Ultra 4 with Live Rewind; and AirPods 5. explainx.ai breaks down specs, pricing, the AI/Siri story, and what a hardware-first CEO signals for builders.

Sep 10

Sep 10

Microsoft reportedly agreed to accept breach liability as part of what's described as the first national AI standard governing AI tools in US schools. explainx.ai covers why a vendor accepting liability directly is unusual, what it likely means for how AI tools get deployed in education, and what other ed-tech AI vendors may now face pressure to match.

Sep 10

Microsoft released ThinkingBox and ThinkingBox-Bench — 507 executable tasks across five business domains that grade agents on the records they actually changed, not the confidence of their final response. The reliability data it surfaces (pass@1 vs pass@20 vs pass^20) is the more useful number.

Sep 5

Microsoft followed July's MAI-Image-2.5-Pro with MAI-Image-2.6 and a companion MAI-Image-2.6-Flash tier — a faster, cheaper sibling that trades a few Elo points for roughly double the throughput. This is the version-bump story, what "Flash" naming means across labs, and when the speed trade is worth taking.

Sep 3

Apple confirmed its "Surprise and Shine" event for September 9, 2026 — the first major product launch under new CEO John Ternus, expected to bring the iPhone 18 Pro and Pro Max, Apple's first foldable iPhone, an Apple Watch Series 12 and Ultra 4, and AirPods 5. Here's everything confirmed versus leaked, plus a look at the leak videos circulating ahead of the keynote.

Sep 1

Tim Cook stepped down as Apple CEO after 15 years on August 31, 2026, handing the role to John Ternus, previously SVP of Hardware Engineering. Cook stays on as executive chairman. explainx.ai looks at what a hardware-first successor means for Apple's still-catching-up AI strategy — Siri's Gemini deal, on-device LLMs, and the privacy-first architecture Ternus inherits.

Aug 21

Aug 21

Aug 21

Aug 21

Aug 20

Copilot Studio is where Microsoft enterprise loops actually live: Workflows that start on a schedule or connector event, autonomous agents with event triggers, and deterministic agent flows. This guide maps each surface with copy-paste instructions and honest limits — separate from GitHub Copilot and Microsoft 365 Copilot chat.

Aug 19

A demo video buried in the macOS Tahoe 26.7 release candidate shows AirPods with a camera in the stem, letting Siri "remember" what you're looking at through a feature called Visual Intelligence. The leak traces back further than the viral tweet suggests — Bloomberg's Mark Gurman flagged the project in May, and Ming-Chi Kuo has tracked it since 2024. Here's what's confirmed, what's speculation, and how this would actually have to work.

Aug 13

Microsoft has opened MAI-Thinking-1 in Foundry public preview. The 1T-total, 35B-active MoE is strong for its size, but it is closed-weight, text-only, and depends on the application for tools and security boundaries.

Aug 10

Microsoft researchers analyzed one week of GitHub Copilot's coding-agent traffic — 761 million LLM calls across 3.2 million users — and found 87% of those calls were fired autonomously by the agent, not typed by a human. explainx.ai breaks down what the paper actually measured, why "Copilot agents" here means GitHub Copilot specifically, and what it implies for where agentic demand is really concentrated in 2026.

Aug 7

A wheresyoured.at analysis of Microsoft's FY26 filings found that roughly 70% of the company's reported AI revenue traces back to OpenAI's own Azure spending — money Microsoft invested flowing back as "growth." Here's how the loop works, what the 20%-revenue-share cap actually caps, and why the GPU depreciation schedule underneath it all is the more consequential fight.

Aug 3

Microsoft’s Generative AI for Beginners remains the default open curriculum for shipping first GenAI apps — 21 Learn/Build lessons, Python + TypeScript, 50+ translations. explainx.ai covers the 2026 Foundry Models modernization and how to start without cloning every translation blob.

Aug 1

After Microsoft’s earnings call, Satya Nadella showed ROIC Intelligence: a Fabric + Copilot + GitHub Enterprise app built from Brian Nowak’s Hyperscale ROIC PDF using /drill-me, autopilot, and /rubber-duck. explainx.ai unpacks the architecture, Agent 365 rails, and the “not tokenmaxxing” thesis.

Jul 29

Microsoft is putting its in-house image and speech models into Foundry public preview: a highest-fidelity image tier for text and localized edits, plus a lower-latency voice tier for high-volume agents. This guide separates documented product evidence from launch claims and shows how to evaluate each path.

Jul 28

Jul 27

Jul 24

Jul 24

Jul 18

Jul 17

Jul 12

Jul 7

Jul 3

Jun 25

Jun 25

Jun 9

Apr 29