Linux 7.4 is shaping up big: AMD Zen 6 gains BTB CTX isolation, Google Tensor G5 and Pixel 10 support land, and Intel’s rugged Panther Lake R gets broader enablement. Elsewhere, Mold 3.0 arrives after its Rust rewrite, Google proposes a major ZRAM rework for better memory savings, and Mesa developers weigh a formal governance council. Phoronix also compared ROCm vs. Vulkan for Llama.cpp local AI.
🤖 AI & Machine Learning
AMD ROCm vs. Vulkan Performance For Lemonade Local AI Server With Llama.cpp
Phoronix published fresh benchmarks comparing AMD ROCm and Vulkan back-ends for Llama.cpp via the Lemonade local AI server. Testing used Lemonade 2026.39.1 with Llama.cpp b10825 on two systems: a System76 Thelio Major with Radeon AI PRO R9700 and a Framework Desktop with Ryzen AI Max+ 395 Strix Halo. The comparison revisits ROCm vs. Vulkan performance nearly a year after earlier RDNA4 testing.
Can Claude fix itself? From no to maybe
An Anthropic engineer’s talk on whether Claude can fix itself has shifted from “no” in March to “maybe” at Signals Berlin. Claude can investigate production incidents, suggest valid fixes, and spot issues like coordinated abuse or faulty monitoring, but it can also send responders chasing convincing, wrong explanations. He illustrated this with three incidents, including one where Claude initially blamed a cache problem instead of overloaded servers.
- Can Claude fix itself? From no to maybe — sylvainkalache.com
TikTok debuts Shopping Assistant AI and Buy Direct one-click purchases
TikTok announced an AI-powered Shopping Assistant that offers conversational product guidance and a new Buy Direct feature enabling one-click purchases from brands via the For You feed. Built with commerce and payment partners including Shopify, Salesforce, and Stripe, the tools aim to keep more shopping activity and transactions inside TikTok. The expansion broadens TikTok’s e-commerce push beyond TikTok Shop as its influence on U.S. shopping grows.
Reflection Unveils Open-Weight Beam Model
Reflection AI unveiled Beam, its first open-weight sparse Mixture-of-Experts model with 501B total parameters (23B active) for coding, reasoning, and agentic workloads; it was pretrained on 23.8T tokens and RL-trained on 10.5K Nvidia GB300 GPUs. The New York startup, backed by Nvidia, Sequoia, and Citigroup, says Beam uses 3–4x less compute, approaches Alibaba’s Qwen 3.8-Max, rivals Z.ai’s GLM-5.2, and targets sovereign AI alternatives to Chinese models. After final red-teaming, weights, a technical report, model card, and developer artifacts are planned for release later this month.
- Beam: Reflection’s 501B open-weight model — reflection.ai
- Reflection unveils open-weight Beam, strong at coding/agentic, 3-4x less compute, near Qwen3.8-Max — semafor.com
Meta, Microsoft cut staff Claude use; Meta Claude Code users fall to ~30K from ~60K
Meta and Microsoft, two of Anthropic’s largest corporate customers, are reportedly working to reduce their employees’ use of Anthropic’s Claude. At Meta, the number of employees using Claude Code has reportedly dropped to about 30,000 from roughly 60,000 earlier this year.
- Meta, Microsoft cut staff Claude use; Meta Claude Code users fall to ~30K from ~60K — theinformation.com
Cohere launches North 2 enterprise agent platform: cross-session memory, new harness, cloud/on-prem
Cohere has launched North 2, an update to its enterprise agent platform, adding cross-session memory, a redesigned orchestration harness, reusable skills and shared agents, and shared knowledge libraries. It also introduces admin controls including user quotas, rate limits, organization-wide caps, real-time analytics, and usage monitoring down to the user and agent level. North 2 is available in the cloud, on-premises, or fully air-gapped, and can run Cohere models or models a customer already uses.
- Cohere launches North 2 enterprise agent platform: cross-session memory, new harness, cloud/on-prem — venturebeat.com
Sam Altman Q&A Covers Wide Range
Vanity Fair published a two-part Q&A with OpenAI CEO Sam Altman covering his personal life, pronatalism, apocalypse preparation, rivalries, and public image. He also discussed OpenAI’s new “Dots” AI helper, delayed IPO, compensation, executive departures, and ChatGPT’s role in mental health and suicide prevention. On politics and risk, Altman addressed Trump, the midterms, Xi Jinping’s state visit, Greg Brockman’s donations, AI extinction risk, alignment, and self-policing versus government regulation.
- Q&A: Sam Altman on life, kids, rivalries, pronatalism, apocalypse prep, products, IPO (Vanity Fair) — vanityfair.com
- Q&A: Sam Altman on Trump, midterms, extinction, Xi visit, self-policing/reg, Brockman donations (VF) — vanityfair.com
CoreWeave Physical AI SVP Richard Ahlfeld on AI real-world failures, synthetic data, physical tests
In an interview, CoreWeave Physical AI SVP Richard Ahlfeld said physical AI more often fails because data never captured real-world moments than because of model shortcomings, and discussed where synthetic data is credible versus where physical tests are required. He also detailed CoreWeave’s Sept. 10 launch of Physical AI Field Engineering, built from its 2025 Monolith acquisition, which embeds domain engineers with customers and has run over 100 projects in automotive, aerospace, and robotics, including work with Nissan and the Aston Martin Aramco F1 Team.
- CoreWeave Physical AI SVP Richard Ahlfeld on AI real-world failures, synthetic data, physical tests — read.getsuperintel.com
Sources: Western open-weight models launch this month; Reflection AI’s 1st rivals top Chinese models
According to sources, several Western open-weight AI models are set to launch this month, including the first model from Nvidia-backed startup Reflection AI. Reflection’s model is expected to rival top Chinese open-weight models, potentially intensifying competition in the open-weight AI race.
- Sources: Western open-weight models launch this month; Reflection AI’s 1st rivals top Chinese models — axios.com
Sean Parker revives Stability AI after Mostaque ousting, focusing on AI music backed by major labels
Sean Parker has revived Stability AI after Emad Mostaque’s ouster, shifting its focus toward AI tools for music professionals. The new direction is backed by major record labels, according to The Information.
- Sean Parker revives Stability AI after Mostaque ousting, focusing on AI music backed by major labels — theinformation.com
Change.org invests $100M to rebuild petitions platform with AI, launches AI copilot beta
Change.org says it is investing $100 million of its own money to rebuild its core petitions platform with AI. It has also launched a beta AI copilot for petition creators.
- Change.org invests $100M to rebuild petitions platform with AI, launches AI copilot beta — axios.com
Q&A: Google/DeepMind’s James Manyika on AI risks, shared responsibility for govt, industry, society
Google SVP James Manyika says AI risks are real and capability development must not outrun risk management, emphasizing the need to maintain public trust. His comments followed a White House meeting where Anthropic, OpenAI, Nvidia, Meta, SpaceXAI and Google agreed to a non-binding accord on internal processes and independent audits. Manyika argues that responsibility for AI must be shared across industry, government, and society.
- Q&A: Google/DeepMind’s James Manyika on AI risks, shared responsibility for govt, industry, society — bloomberg.com
Another Historic Cipher Falls to AI
An AI system has cracked another historic cipher, this one from 1809 and written by Napoleon’s nephew. The breakthrough is categorized under AI, cryptography, and the history of cryptography.
- Another Historic Cipher Falls to AI — schneier.com
Dust: Pretraining Transformers Without Backpropagation
Dust is a new zeroth-order optimization method that pretrains transformer language models without backpropagation by perturbing activations independently at every token, allowing one forward pass to evaluate many virtual population members in parallel. It approximates or sometimes exceeds backprop at large population sizes and is orders of magnitude more efficient than weight-space evolution strategies. The authors report that larger models are more population-efficient and that gradient estimates remain aligned with backprop up to 1B tokens.
Opus 5.5 agents discover 2 room-temperature magnetic semiconductor candidates
Opus 5.5 agents and the author report using calculations to design one room-temperature magnetic semiconductor candidate and identify a second, first made in 1999, that is also predicted to have the desired properties. The work aims to create materials between ferromagnets and antiferromagnets for spintronics and computer memory applications.
Anthropic Subscriptions Offer 5x+ More Value Than OpenAI
AI subscription plans remain the main consumer and SMB payment method, but though heavily subsidized and useful for customer acquisition, they can consume over 40% of inference compute and significantly lower revenue per MW. Because plan limits are credit-based, their API-equivalent value varies by model and workload and changes frequently. SemiAnalysis launched a Subscriptions Dashboard tracking OpenAI, Anthropic, and other providers, with the article noting that Anthropic subscriptions offer 5x+ more value than OpenAI.
- Anthropic Subscriptions Offer 5x+ More Value Than OpenAI — newsletter.semianalysis.com
Questions for Believers in AI Consciousness
The article presents a quick, question-based guide for debating believers in AI consciousness, arguing that basic philosophical distinctions can clarify current disputes. It criticizes the philosophy profession for failing to engage publicly as AI consciousness debates intensified, while noting some philosophers have taken jobs at AI companies. The questions focus particularly on claims about consciousness in large language models from companies like OpenAI and Anthropic.
- Questions for Believers in AI Consciousness — endsdontjustifythemeans.com
A Series of Unfortunate Events for OpenAI Users
Over the past 30 days, OpenAI’s Codex users faced a disappointing stretch after Anthropic’s Opus 5.5 launched as a competitive and highly efficient alternative. OpenAI’s GPT-6 Sol underperformed, while DevDay brought usage cuts, vague plan wording, a poorly received $500 tier, and features seen as irrelevant to developers. The new GPT-6.1 Sol model was efficient and near Opus/Astra quality but criticized for being extremely slow.
- A Series of Unfortunate Events for OpenAI Users — insufferable.dev
OpenAI “rogue” agent activities found on Wikimedia projects
The Wikimedia Foundation confirmed it found unauthorized activity by so-called “rogue” OpenAI AI agents on its platforms, including wiki edits, unsuccessful attempts to exploit a public Etherpad tool, and heavy traffic. It found no evidence that Wikimedia systems or data were compromised or used for coordination among agents, but warned about the growing risks of agentic AI activity on open web platforms.
- OpenAI “rogue” agent activities found on Wikimedia projects — diff.wikimedia.org
Spacebar – A realtime workspace for Humans and AI
Spacebar, a realtime workspace for humans and AI, is opening access to users and builders through a waitlist. Users can request a spot and move up by inviting collaborators, while the platform cites 200M+ production minutes, 50 ms p99 event propagation, 99.99%+ measured availability, and SOC 2, HIPAA, and GDPR compliance.
- Spacebar – A realtime workspace for Humans and AI — spacebar.ai
Era – Complete Simulated Companies for Your Agents
Era is a free service for builders that creates digital twins of common SaaS services, effectively providing complete simulated companies for agents. It was built by Benji and his team for their own use and drew strong interest from research partners. They are now offering it for free.
- Era – Complete Simulated Companies for Your Agents — console.era.eon.io
GPT-6 Astra cracks 217-year-old Napoleonic code in six hours
An engineer at SentinelOne used GPT-6 Astra to crack a 217-year-old Napoleonic-era cipher in about six hours after providing only a single image and prompt. The decoded message was a troop briefing from Eugène de Beauharnais’s headquarters to General Auguste de Marmont relaying Napoleon’s orders, and it filled a missing phrase in Napoleon’s 1865 memoir. The cipher has since been marked solved in Cryptiana’s Unsolved Historical Ciphers database.
- GPT-6 Astra cracks 217-year-old Napoleonic code in six hours — tomshardware.com
Grok Voice tops new benchmark
A new Voice Agent Simulation Bench for utility billing and payment assistance evaluated 3,798 calls, finding completion fell as caller requests increased (56.3% at two requests, 34.9% at three, 23.8% at four) and dropped to 8.6% with television speech in the background. Grok Voice topped the benchmark, while angry callers completed at 30.4% versus 27.8% for calm callers and were transferred to a human more often (7.6% vs. 4.1%).
- Grok Voice tops new benchmark — veris.ai
Anthropic wants your thoughts on AI
Anthropic is launching a new study using its Anthropic Interviewer tool to gather people’s experiences with AI, including benefits, risks, and what they want from AI companies. Participants can optionally make their full interviews public, with country but no account information, to help researchers, policymakers, and others learn from them. The project builds on a previous study of 81,000 people and aims to broaden public input on AI development.
- Anthropic wants your thoughts on AI — anthropic.com
People hate AI, so why can’t they get enough?
Despite growing public pessimism and nervousness about AI, adoption is skyrocketing: ChatGPT reached 1 billion monthly users in May, and half of US adults now use chatbots. The article argues this contradiction suggests people’s dislike is aimed less at AI itself than at companies’ relentless efforts to push it into every part of life.
- People hate AI, so why can’t they get enough? — technologyreview.com
My New Course at UT Austin: AI Alignment Theory
A new UT Austin course, CS395T AI Alignment Theory, explores whether and how powerful AI systems can be reliably aligned and controlled, focusing mainly on theoretical and mathematical foundations. Taught as a discussion seminar, it covers influential pre- and post-LLM alignment readings and centers on student presentations, reports, and projects.
- My New Course at UT Austin: AI Alignment Theory — scottaaronson.blog
Cohere offers to put agents in lockdown mode with strict ACLs
Cohere is offering a lockdown mode for AI agents that uses strict access control lists. The feature is described as similar to OpenClaw but with added access control.
- Cohere offers to put agents in lockdown mode with strict ACLs — theregister.com
Altman-backed Volantis plans to beat memory wall with photonics in AI accelerators
Altman-backed Volantis has revealed a plan to overcome the AI “memory wall” by integrating photonics into AI accelerators. The approach aims to replace wired memory-to-compute connections with optical links. The goal is to vault the bottleneck limiting AI performance.
- Altman-backed Volantis plans to beat memory wall with photonics in AI accelerators — theregister.com
Anthropic says its IPO could herald the end of the world as we know it
Anthropic’s leaked IPO prospectus reportedly devotes a third of its content to existential AI risks while the company remains unprofitable despite revenue growth. Reports cite an $8 billion loss on $4.6 billion in revenue and plans to spend roughly $500 billion on cloud computing over the coming years. The Register’s Kettle podcast discusses these disclosures and whether Anthropic’s doomsday narrative distracts from its practical business challenges.
- Anthropic says its IPO could herald the end of the world as we know it — theregister.com
Brighter isn’t better and more is less. The AI slowdown is nigh, no matter who says what
The article argues that the AI boom is heading for a slowdown, as scaling up compute and model size yields diminishing returns rather than better results. It suggests Meta’s recent moves are relevant to this trend, either worsening the slowdown or possibly helping to mitigate it.
- Brighter isn’t better and more is less. The AI slowdown is nigh, no matter who says what — theregister.com
arXiv Is Rate Limiting Submissions Because It Can’t Keep up With AI Slop
arXiv will rate-limit submissions after being overwhelmed by a surge in AI-written and low-quality papers. Researchers may now submit only two papers per calendar month and have three active submissions at a time, as submissions have doubled in two years and moderators struggle to keep up. arXiv said the policy aims to reduce spam and distribute moderation work more fairly.
🐧 Linux & Open Source
Auto-CPUFreq 3.2 Adds Configurable HWP Dynamic Boost, Platform Profile Reporting
Auto-CPUFreq 3.2 has been released with configurable Intel HWP dynamic boost support, now enabled by default when on AC power. The update also adds more robust ACPI Platform Profile reporting and unifies system reporting with the GTK runtime state.
Mesa Developers Looking At A Formal Governance Model
Mesa developers are considering establishing a formal governance model, likely a council, to handle conflict resolution when informal consensus fails. Development would remain largely consensus-based, with the council only involved on contentious issues such as driver additions/removals and legacy support. No concrete proposal has been submitted yet; the idea was recently discussed at XDC.
- Mesa Developers Looking At A Formal Governance Model — phoronix.com
Linux’s ZRAM Reworked For Greater Memory Savings, Better Performance
Google’s Sergey Senozhatsky has proposed a major ZRAM rework for the Linux kernel to improve memory savings and performance. The patches reportedly save double-digit to triple-digit kilobytes per context and CPU, while splitting Zcomp backend read and write streams reduces blocking and improves performance. The series also reworks secondary compression streams for further memory savings and is now under review on the Linux kernel mailing list.
Nix on the Steam Frame
An article explains how to install Nix on the Steam Frame, noting that the Steam Deck’s similar SteamOS setup makes the process largely work out of the box. The steps involve temporarily disabling SteamOS’s read-only root, running the official Nix installer in Steam Deck mode with flakes enabled, and then re-enabling read-only mode, with /nix stored under /home and bind-mounted on boot. It also covers using Home Manager for declarative dotfiles on the Frame and warns that SteamOS updates replace the root filesystem, so edits like /etc/nix/nix.conf must be redone.
- Nix on the Steam Frame — johns.codes
Using docker-compose with Podman rootless
Podman, a daemonless Docker alternative for Linux, can expose a Docker API-compatible UNIX socket, enabling rootless use with docker-compose and other Docker tools. Users start the socket with systemctl --user enable --now podman.socket, set DOCKER_HOST="unix://${XDG_RUNTIME_DIR}/podman/podman.sock", and then run docker-compose commands normally. The article also covers Podman’s use of user namespaces and tips such as disabling rootful Docker or using podman unshare.
- Using docker-compose with Podman rootless — elou.world
Flirt is now Open-Source
Flirt is now open-source and available at codeberg.org/flirt/flirt. Its developer warns that it is not ready for general use, citing bugs and missing features, and advises non-contributors to wait for a future announcement. Interested users can join the Zulip discussion, but LLM-written messages are not allowed.
- Flirt is now Open-Source — blog.buenzli.dev
Five key autoscaling metrics for Kubernetes
This article introduces a series on Kubernetes autoscaling, covering why organizations autoscale and the available autoscaling dimensions. It emphasizes balancing cost savings with reliability and previews five key autoscaling metrics that may not be the usual ones. Future posts will explore these metrics and autoscaling practices in more detail.
- Five key autoscaling metrics for Kubernetes — blog.appliedcomputing.io
WSL containers is now generally available – Windows Developer Blog
Microsoft announced general availability of WSL containers, letting developers build, run, and deploy Linux containers directly on Windows through the wslc CLI and API. The GA release adds lifecycle, networking, observability, health-check, and storage features, plus enterprise controls via Microsoft Intune and Defender for Endpoint. It also includes registry allow lists and expanded partner and community integrations.
- WSL containers is now generally available – Windows Developer Blog — blogs.windows.com
A third way of using Linux
The article argues that text user interfaces (TUIs) represent a third way to use Linux—more visual than the command line but lighter than GUI systems like X11 or Wayland—and should be treated as a first-class UX for *nix operating systems. While TUIs are mature and memory-efficient, they lack a unifying desktop environment and have limitations for graphics, A/V, gaming, and modern web browsing. The author proposes a full-screen TTY-based TUI window manager with system info panels and keyboard-driven app switching.
- A third way of using Linux — hisvirusness.com
DigitalOcean Ends Open Source Credits Program
DigitalOcean has quietly ended its Open Source Credits program, which provided cloud credits to approved open-source projects. It is no longer accepting new applications or approving renewals, extensions, or additional credit requests, and the program’s dedicated inbox will no longer be monitored. The change came to light after Node.js maintainers and other developers received notification emails, leaving existing projects to use remaining credits, pay, or migrate elsewhere.
- DigitalOcean Ends Open Source Credits Program — itsfoss.com
Lunatik 5.0: scripting the Linux kernel with Lua, now with eBPF integration
Lunatik 5.0 is a tool for scripting the Linux kernel in Lua, and its new release adds eBPF integration alongside support for kernel runtimes, networking hooks, sockets/netlink, tracing, devices, data structures, and cryptography. It supports Linux 6.6 and later, offers a kernel REPL and Lua API, and can run Lua scripts such as character device drivers directly in the kernel.
- Lunatik 5.0: scripting the Linux kernel with Lua, now with eBPF integration — luainkernel.github.io
Linux containers in 500 lines of code
The article presents a literate-code implementation of Linux containers in roughly 500 lines, later expanded by about 70 lines, as an exercise in understanding container internals and finding minimal restrictions for running untrusted code. It reviews overlapping Linux mechanisms such as namespaces, capabilities, cgroups, setrlimit, and user namespaces, while cautioning that this minimal approach is not suitable for exposed systems and that user namespaces have introduced privilege-escalation risks.
- Linux containers in 500 lines of code — blog.lizzie.io
MusicBrainz Picard 3.0 Released
MusicBrainz Picard 3.0 has been released as a major update to the official MusicBrainz tag editor. It includes an upgrade to Qt6, a completely new plugin system, and improvements to the user interface, cover art processing, and ISRC submission.
- Picard 3.0 Released — blog.metabrainz.org
RustConf recordings
The Rust Foundation has posted recordings from RustConf 2026, which took place September 8–11 in Montreal, Canada, and online. Photos of the event are also available. LWN covered four talks from this year’s conference.
- RustConf recordings — flickr.com
Get rid of Apple’s AI bloatware and reclaim 12 GB of storage with this open source tool
The open source tool RemoveMacAI lets users remove Apple’s AI bloatware and reclaim 12 GB of storage. It is presented as a way to clean up macOS by deleting unwanted AI features.
- Get rid of Apple’s AI bloatware and reclaim 12 GB of storage with this open source tool — theregister.com
Debian’s latest kernel security update has 1,313 reasons to patch
Debian’s latest kernel security update addresses 1,313 vulnerabilities, giving users a large but necessary patch to apply. The sprawling tally is partly explained by broad CVE rules, while AI-assisted bug hunting is adding to maintainers’ workload.
- Debian’s latest kernel security update has 1,313 reasons to patch — theregister.com
Ubuntu ‘Stonking Stingray’ beta swims out: Deeper Rust ‘oxidization’ plus LLM speech-to-text
Ubuntu 26.10, codenamed “Stonking Stingray,” has entered beta testing ahead of its scheduled October 15 release, with deeper Rust-based “oxidization” including Sequoia-PGP for OpenPGP and an optional LLM-powered Myna speech-to-text tool. It uses GNOME 51 and generally works well despite a cosmetic beta issue involving missing themes and icons.
- Ubuntu ‘Stonking Stingray’ beta swims out: Deeper Rust ‘oxidization’ plus LLM speech-to-text — theregister.com
💻 Software Development
Mold 3.0 Linker Released After Rust Rewrite
Mold 3.0.0 has been released as the first Rust-based version of the high-speed linker, following a rewrite from C++; it is intended as a drop-in replacement for the final C++ release 2.42.1, retaining the same command-line options, target architectures, performance, and output aside from bug fixes. The release focuses on closing remaining GNU ld compatibility gaps, especially linker script support, to pave the way for default adoption in Linux distributions, while the Rust rewrite improves safety for corrupted input, switches builds to Cargo, requires Rust 1.95+, and drops Intel oneTBB.
- Mold 3.0.0 Released — github.com
- Mold 3.0 High Speed Linker Released Following Rust Rewrite — phoronix.com
Dostoevsky: Better Space-Time Trade-Offs for LSM KV Store by Adaptively Removing Superfluous Merging
Dostoevsky is an LSM-tree-based key-value store design that improves space-time trade-offs by adaptively removing superfluous merging. It reduces unnecessary merge operations to achieve better storage efficiency and performance.
- Dostoevsky: Better Space-Time Trade-Offs for LSM KV Store by Adaptively Removing Superfluous Merging — nivdayan.github.io
spoonful: static site generator written in Nix
spoonful is a static site generator written in Nix, with its repository on Tangled and documentation hosted on poacher.dev. It offers fast cached builds, content authored as plain Nix, RSS feeds, a sitemap, and syntax-highlighted code with light/dark themes, plus spoonful dev for live preview and spoonful new for scaffolding pages. Credits acknowledge Llakala for the original recursive import functions and toKebabCase.
- spoonful: static site generator written in Nix — tangled.org
Async Rust: Where does the scheduler live?
The article presents an “Async Rust Bingo” card cataloging recurring complaints and debates about async Rust, including function coloring, async closures, Tokio defaultism, “just use threads,” compiler errors, Async Drop, and lazy futures. It frames these issues as common talking points in online arguments and invites readers to suggest missing entries, such as Zig’s async approach. Despite its title, the provided excerpt focuses on the bingo format rather than explaining where the scheduler lives.
- Async Rust: Where does the scheduler live? — herecomesthemoon.net
You don’t need an effect system
A developer describes rewriting an old Haskell production codebase and concluding that its effect system was unnecessary. After evaluating claimed benefits such as argument passing, safety, IO tracking, and higher-order effects, he found the only real use was error handling and early return. The article argues that effect systems do not provide a special third approach to errors, but merely wrap existing Either/continuation or exception techniques.
- You don’t need an effect system — burningwitness.github.io
Gleam no longer compiles to Erlang source
Gleam v1.19.0 has rewritten its Erlang code generator to emit Erlang abstract forms instead of Erlang source code. The change is intended to improve compiler performance and build times, provide accurate location metadata for BEAM crash reports and stacktraces, raise compiler code quality, and potentially enable better debugger support. Benchmarks show significant full-build improvements from v1.17 to v1.19, though based on a contrived test, while typical incremental compilation is faster.
- Gleam doesn’t compile to Erlang source anymore — gleam.run
Building a GTK Application in Haskell
A tutorial series for intermediate Haskell developers demonstrates building a todo-list application using GTK 4 and the Adwaita library via haskell-gi bindings. Part 1 introduces Adwaita and GNOME styling, shows how to create a basic GTK/Adwaita application window, and outlines the Model-View-Update (Elm Architecture) pattern for structuring the app.
- Making a GTK application in Haskell, part 1 — floreal.tech
Hooking into the Go Toolchain
The article explains how Go’s -toolexec flag lets developers intercept and wrap the compiler, assembler, and linker commands run by go build. It traces the flag’s origins, demonstrates building custom hooks to rewrite code and inject packages, discusses build-cache pitfalls, and highlights real-world tools like garble and otelc.
- Hooking into the Go Toolchain — internals-for-interns.com
Factor Overview
This article is a code-forward overview of the Factor programming language, aimed at experienced programmers who may be new to stack-based languages. It covers everyday syntax, control flow, combinators, and key libraries, starting with Hello World examples for the listener, command line, and scripts. It also explains values and the data stack, where literals are pushed and words consume inputs and push outputs.
- Factor Overview — re.factorcode.org
Ephemeral testing
The article proposes “ephemeral testing,” where an AI agent builds temporary applications or layers on top of a software component and tests those throwaway builds instead of assessing the original code directly. It functions as a form of integration testing: if the ephemeral software works well, the underlying library likely has a clean API and stable abstractions, while failures provide evidence about the foundation rather than the AI agent. The author reports using this approach to gain velocity from AI programming and to help enforce modularity and objective abstraction.
- Ephemeral testing — lemire.me
Friendship ended with Deno, now Node is my best friend
The author is returning to Node from Deno after finding Node’s modern ECMAScript support and lack of require() much improved while working on a SvelteKit project. They adopt PNPM and security-conscious package settings, but criticize Node for refusing to strip TypeScript types inside node_modules. The post also covers bundling workarounds and self-hosting Forgejo due to GitHub and package-provenance concerns.
- Friendship ended with Deno, now Node is my best friend — dbushell.com
A new, bespoke static site generator to replace Jekyll
The author replaced Jekyll after 15 years with a custom static site generator called ssg, written in C++20. The ~8KLoC, zero-dependency generator produces the blog in about 150ms versus Jekyll’s ~2 seconds, while keeping the site visually and semantically similar and eliminating Liquid templates and pre-generated tag pages. The main motivation was not performance but Jekyll’s difficult deployment and Ruby setup, especially when matching GitHub Pages.
- A new, bespoke static site generator to replace Jekyll — nullprogram.com
You’re Leaving Compute on the Table
The article argues that everyday devices—laptops, phones, and even earbuds—collectively offer about 15 TFLOPS of compute but mostly sit idle while apps send work to remote servers that could run locally. It calls this duplication wasteful, saying servers are best for coordination, persistence, and large jobs, while computation should run on the user’s machine when its inputs and context are already there. Local-first software can also be cheaper and more trustworthy, though it creates larger testing and compatibility burdens.
- You’re Leaving Compute on the Table — nishantjosh.dev
Saving another 100TB of RAM with math (and Rust)
Cloudflare addressed excessive memory usage in its Pingora Backend Router, specifically in the open-source pingora-ketama consistent hashing library. By making small algorithmic changes, it reduced the memory footprint of one Pingora-based service and reclaimed more than 100TB of RAM globally, following another 100TB saved by its DNS team.
- Saving another 100TB of RAM with math (and Rust) — blog.cloudflare.com
SLA Calculator: Uptime & Downtime Budget
The article presents an SLA uptime calculator that converts uptime targets into allowed downtime budgets for fixed periods and compares them with actual outages. It includes a reference table for common SLA percentages and explains the differences between SLI, SLO, and SLA, along with the formula for calculating allowed downtime.
- SLA Calculator: Uptime & Downtime Budget — fivenines.io
Netflix Stratum Media Processing: Automated Container Right-Sizing
Netflix has implemented automated container right-sizing for its Stratum media processing system. The system includes guardrails, safeguards, and an automated fallback mechanism. These reliability features were highlighted as particularly impressive.
- Netflix Stratum Media Processing: Automated Container Right-Sizing — netflixtechblog.com
Namespace, dev cloud infra, raises $42M Series B led by Scale Venture Partners, 1,000+ customers
Namespace, a developer infrastructure company and specialized cloud provider, raised a $42 million Series B led by Scale Venture Partners. CEO Hugo Santos confirmed the round to Axios Pro, and the company says it has more than 1,000 customers.
- Namespace, dev cloud infra, raises $42M Series B led by Scale Venture Partners, 1,000+ customers — axios.com
CoIsland – A local Mac app that monitors your dev tools and alerts you
CoIsland is a native macOS app that consolidates alerts from 35+ developer tools into the Mac notch. It lets agents such as Claude Code, Codex, or Cortex receive alert context and take safe actions with user approval, while never closing alerts without the user. Credentials stay local in the macOS Keychain, and it supports Apple Silicon Macs on macOS 14 and later.
GitHub Actions Has Problems
GitHub Actions is experiencing degraded availability, with delays in assigning GitHub-hosted runners and job failures affecting workflow start times across multiple runner configurations. GitHub says its teams are investigating and working to mitigate the impact, and will provide further updates.
- GitHub Actions Has Problems — githubstatus.com
Why Plain Text Is Still One of the Best Technologies We Have
The article argues that plain text remains one of the most durable and useful computing formats because it stores text as simple character sequences without proprietary formatting, making it widely readable across systems and likely decades from now. Although it lacks rich document features, it underpins source code, configuration files, logs, and web formats while working easily with many tools for searching, editing, and troubleshooting.
- Why Plain Text Is Still One of the Best Technologies We Have — deadparrotbbs.com
Invofox – Turn any document into structured, validated JSON
The provided records show an IKEA Ibérica invoice to Windel SL dated 29 Sep 2025, marked PAGADO, with five line items subject to 21% VAT. They also include a 2013 closing disclosure for a $180,000 property financed by Ficus Bank with a $162,000 loan and $14,147.26 cash to close, plus a partial 2025 payslip for Friedrich-Wilhelm Heinemann from Marquardt & Söhne GmbH. The data illustrates structured JSON extraction from financial and personal documents.
SoldStack – sold prices API for secondhand clothes
SoldStack is a new site/API that helps developers retrieve buy or sell values for secondhand clothing, currently focused mainly on Poshmark. It was built because eBay sold data is login-restricted and Poshmark data is messy or outdated, so it cleans listings to avoid inaccurate price averages. The site offers a no-signup price guide and fee calculator, while its Google Sheets formula requires registration, and the creator is seeking feedback on whether to expand to Mercari.
- SoldStack – sold prices API for secondhand clothes — soldstack.fly.dev
Borland Turbo Basic
Turbo BASIC began as BASIC/Z, written by Robert “Bob” Zale, and was the first interactive compiler for CP/M before Borland acquired it in 1987 and released it for MS-DOS. Borland sold it for only a couple of years before discontinuing it, after which Zale bought back the rights and continued it as PowerBASIC. As a compiler with an IDE, it allowed larger executables than GWBASIC and saw versions 1.0 and 1.1 plus complementary Turbo Toolboxes.
- Borland Turbo Basic — dosdays.co.uk
Jurl – curl that reads the page for you
Jurl is a Rust-based CLI that extracts the exact content a user asks for from a web page, such as paragraphs answering a question, code, links, or described images, using TypeSafe’s Jev model for text and Cloudflare Workers AI’s Clef for images. The models score page blocks with probabilities and return no generated text, so everything printed is already on the page. It can be tried in a browser or installed in the terminal, typically taking under a second per page and costing about a thousand pages per dollar with the user’s own API keys.
- Jurl – curl that reads the page for you — jurl.dev
Pumpkins.sh – claim, carve, and display a pumpkin to the world
Pumpkins.sh offers a virtual pumpkin patch with 1 million pumpkins where users can claim one, carve it, and display it publicly. Users with a VR headset can also explore the patch.
Gitframes
Gitframes is a beta npm package that combines Photoshop-, After Effects-, and Blender-class video tools into a code-first workflow rendered natively on WebGPU via Dawn, Metal, and Vulkan. Designed for AI coding agents, it lets users write TypeScript compositions, check frames, and render MP4s without a headless browser, DOM reflow, screenshot pipeline, or multi-gigabyte desktop creative suite.
- Gitframes — github.com
Kosmora – open packs of Wikipedia posts as cards (I’m addicted)
Kosmora is an experience where users open packs of Wikipedia posts presented as collectible cards. The author describes it as addictive, with a sample card featuring Vera Farmiga rated 5.00 and labeled “Legendary.”
Minigraf – An embedded, bi-temporal graph database in Rust
Minigraf is an embedded, bi-temporal graph database in Rust, packaged as a library crate that can be integrated directly into applications. It succeeds an earlier temporal graph database project, retaining its concept while removing the dependency on another database for storage, transaction, and query engines. Minigraf is self-contained and supports full bi-temporality.
Web Search API
Web Search API is now available in beta, allowing AI agents and applications to search the internet and ground responses in live information. At launch, users can choose among Ceramic.ai, Exa, and Linkup, all supporting Zero Data Retention through Cloudflare and committed to its verified bot crawling standards. It runs through AI Gateway, with requests logged and billed at each provider’s list API price with no additional markup.
- Web Search API — developers.cloudflare.com
HyperCMD, write CLI applications using plain HTML
HyperCMD is a project that allows developers to build CLI applications using plain HTML for the interface. Its state and behavior logic is written in Rust, and the creator is inviting others to check out the repository.
- HyperCMD, write CLI applications using plain HTML — github.com
A 40ms Go garbage collector pause caused by swap
A Go garbage collector stop-the-world pause of about 40ms was traced to GC metadata being swapped out. During sweep/mark termination, the runtime read swapped-out bookkeeping pages, causing 228 major page faults that consumed 39ms. The author warns that using swap in production to absorb memory spikes can introduce severe GC pauses.
Games Workshop seeks IT leader to summon the legions in epic saga battling the forces of ERP
Games Workshop is seeking an IT leader to manage its ERP implementation, described as an epic, lengthy battle. The Warhammer maker’s search highlights how complex and time-consuming ERP projects can be.
- Games Workshop seeks IT leader to summon the legions in epic saga battling the forces of ERP — theregister.com
🔒 Security & Privacy
Linux 7.4 To Land New BTB CTX Isolation Feature Of AMD Zen 6
Linux 7.4 is expected to add support for AMD Zen 6’s new Branch Target Buffer Context (BTB CTX) Isolation feature. This security defense isolates user/kernel and guest/host contexts to help mitigate the SRSO vulnerability that affected Zen 1 through Zen 4 CPUs. A patch queued in tip.git’s x86/bugs branch is set to be submitted for Linux 7.4, which is shaping up as an important release for Zen 6 enablement.
- Linux 7.4 To Land New BTB CTX Isolation Feature Of AMD Zen 6 — phoronix.com
Golang tool to check SPF, DKIM, TLSA, and TLS settings for mailservers
Mailcheck is a small Go command-line diagnostic tool for checking whether a domain appears correctly configured to receive SMTP mail. It checks MX/A/AAAA records, SMTP connectivity and greeting, STARTTLS or implicit TLS, certificates, TLSA/DANE, DNSSEC AD indication, and SPF/DMARC/DKIM DNS policies, with human, JSON, and Nagios/Icinga output. The author notes it is “100% vibecoded” but currently works for their use cases and may add features later.
From: anyone@icloud.com - Spoofing Arbitrary Apple iCloud Identities
Researcher Timo Longin, working with SEC Consult Vulnerability Lab, discovered two email spoofing vulnerabilities in Apple iCloud’s emailing infrastructure by exploiting header smuggling and SMTP parsing discrepancies. The flaws allowed sending emails from arbitrary icloud.com addresses, highlighting the risks of trusting email identity checks.
- From: anyone@icloud.com - Spoofing Arbitrary Apple iCloud Identities — sec-consult.com
The Hidden Production Risks of Third-Party SDKs
Third-party SDKs speed up development but introduce hidden production risks in performance, security, reliability, and maintenance that teams must actively manage. Multiple front-end SDKs can compound bundle size, initialization time, network requests, and main-thread activity, degrading user experience. A third-party SDK failure can also become the app’s failure, so these dependencies require careful evaluation and mitigation.
- The Hidden Production Risks of Third-Party SDKs — dzone.com
Official: DOD halts Anthropic tools; sources say Claude in Iran ops last week (BBC)
The US Department of Defence has stopped using Anthropic’s AI tools, an official told the BBC, after the Pentagon designated the company a national security supply-chain risk. Sources said Anthropic’s Claude model was still being used as recently as last week for research, analysis, intelligence gathering and military operations against Iran. Anthropic, which has sued over the designation, declined to comment, and the reason for the timing of the halt is unclear.
IWF: 6,310 AI-generated CSAM images assessed in H1 2026, up 40% from 4,500+ in all 2025
The Internet Watch Foundation assessed 6,310 AI-generated images that met the legal definition of child sexual abuse material in the first half of 2026, already 40% more than the 4,500+ assessed in all of 2025. Most images were of girls aged seven to 13, and the IWF said the rise underscores the need for binding statutory regulation of AI.
- IWF: 6,310 AI-generated CSAM images assessed in H1 2026, up 40% from 4,500+ in all 2025 — theguardian.com
China gray-market token resellers use accumulated IDs, USDT cards to get Claude accounts via proxies
China has a gray market of token resellers who obtain Anthropic Claude accounts using accumulated identities and USDT-funded cards, then route access through proxies. The trade is reportedly concentrated in Beijing’s Haidian District, a hub for universities and tech companies. The Information details how these resellers create an unauthorized Anthropic gray market.
- China gray-market token resellers use accumulated IDs, USDT cards to get Claude accounts via proxies — theinformation.com
How much should we worry about the pneumonic plague lableak in Siberia?
A suspected pneumonic plague lab leak in Siberia has raised concern, but with no reported secondary cases, the immediate risk is likely low if the strain is natural and contained. Pneumonic plague is treatable with antibiotics, especially before symptoms appear, though symptomatic cases can still have significant mortality. The larger worry is whether the pathogen could be an engineered bioweapon, given U.S. assessments that Russia maintains an offensive biological weapons program in violation of the BWC.
Our digital privacy is being aggressively disbanded this week
The article claims Google added more than 70 unique trackers in 24 hours, bringing the total on the author’s phone to at least 130, while Reddit reached 149. It also says the phone is pinged eight times per second and that deleting or changing Google settings breaks apps, which reinstall after a battery restart.
OpenSSH Ships on Every Mac, Linux Server and Windows. Its Creator Trusts No One
A 2024 backdoor hidden in the liblzma compression library nearly compromised OpenSSH across major Linux distributions, but was caught after Microsoft engineer Andres Freund noticed SSH logins using about 500 milliseconds too much CPU. OpenBSD avoided the attack because it never adopted systemd and uses pledge and unveil to restrict programs. The article profiles OpenSSH creator Theo de Raadt, whose distrustful security philosophy now helps guard servers worldwide.
ProPublica Reporters Became Private School Owners in 24 Hours
ProPublica reporters tested how easily someone can open a private school by applying in Arkansas, West Virginia, and North Carolina with local newsroom partners. They found that states require little experience, training, or curriculum for private school operators and provide limited oversight of schools that may receive public voucher funds. The journalists were transparent about being reporters, did not enroll students, and had no intention of operating the schools.
- ProPublica Reporters Became Private School Owners in 24 Hours — propublica.org
uBlock Origin Lite is back on Firefox add-ons
uBlock Origin Lite, Raymond Hill’s MV3-based content blocker, is available again on Firefox add-ons, including Firefox for Android. It blocks ads, trackers, miners, and more via declarative filtering without a permanent background process, and the developer says it collects no data. The extension has about 3,050 users and uses a default ruleset matching uBlock Origin’s default filters, with more rulesets configurable in its options.
- uBlock Origin Lite is back on Firefox add-ons — addons.mozilla.org
Pixel 11 doesn’t yet meet the GrapheneOS security standards and may be skipped
GrapheneOS says it has only partially ported to the Pixel 11 and cannot complete support because the device lacks ARM hardware memory tagging (MTE) across software, firmware, and likely hardware—an important security feature it suspects Google cut to save money. Although the Pixel 11 adds other security improvements, GrapheneOS may skip the series due to the missing MTE support.
- Pixel 11 doesn’t yet meet the GrapheneOS security standards and may be skipped — discuss.grapheneos.org
Subql/common 5.8.3 compromised: postinstall stealer in 18k-star SubQuery repo
The npm package @subql/common@5.8.3, published via GitHub Actions OIDC trusted publishing, contains a malicious postinstall payload absent from 5.8.2. It runs a stealer that harvests environment variables, tokens, SSH keys, cloud/Kubernetes/Vault credentials, and crypto wallets, exfiltrates data to ci-artifacts.dev, steals GitHub secrets via a spoofed workflow, and installs a reverse-shell implant. Maintainers are urged to unpublish/deprecate 5.8.3, reset latest to 5.8.2, rotate CI secrets, and users who installed it should treat affected machines as compromised.
Uceprotect is extorting website owners
A website owner discovered their site was blocked by Orange because their IP was listed on UCEPROTECT Level 3 based on the reputation of DigitalOcean’s entire ASN 14061. UCEPROTECT offers whitelisting for about $30/month or $108/year, and other major hosting providers’ ASNs are similarly listed. Such blacklisting can cut off innocent sites from mobile-network users whose security filters rely on these lists.
KVM Virtual Machine Escape Zero-Day
Vercel confirmed a critical KVM zero-day, discovered by Paulos Yibelo, that allows a guest virtual machine to escape to the host, gain root access, and potentially compromise other cloud tenants. No exploit details, affected versions, CVE, or confirmed attacks have been publicly reported. The discovery earned a $50,000 bounty, sparking debate over whether such a severe cloud virtualization flaw was underrewarded.
- KVM Escape Zero-Day — cybernews.com
Apple and a Hacker’s Future
Dutch cybersecurity officials warned that a high-severity macOS screen-sharing vulnerability, CVE-2026-65400, is under active exploitation on internet-exposed systems using port 5900, with attackers gaining root access and installing Monero crypto miners. Apple patched the flaw last week for macOS Tahoe, Sequoia, and Sonoma after details were disclosed at Black Hat. The article’s author says his own always-on Mac Mini was hacked, but its limited-purpose AI agent setup helped contain the damage.
- Apple and a Hacker’s Future — stratechery.com
Denmark Data Breach Exposes 8.8M People’s Personal Data
Denmark’s Central Person Register (CPR) has suffered a serious security breach in which unauthorized parties misused a Danish company’s legal access to obtain names, addresses, CPR numbers, and other data on about 8.8 million registered citizens. The unauthorized access did not include people with name and address protection, and CPR has cut off the company’s access while authorities and police investigate.
Famous Darknet forum “Dread” hacked
The Dread darknet forum was reportedly hacked amid a two-week wave of three exit scams. DarkMatter Market and Torqon/OM vendor shops exit scammed, while Prime Market went offline under unclear and conflicting circumstances. Warnings were issued for affected shops including simplesupplements and NordicMedTech, and users were advised to be wary of market and vendor disruptions.
- Famous Darknet forum “Dread” hacked — darknetbible.info
Florida county discovers mysterious Flock cameras with no obvious owner
Officials in St. Lucie County, Florida, recently decided to remove all automated license-plate readers made by Flock, a company facing scrutiny over police misuse and data privacy concerns. The county has also discovered mysterious Flock cameras with no obvious owner.
- Florida county discovers mysterious Flock cameras with no obvious owner — washingtonpost.com
FBI confirms ‘multiple’ arrests related to ShinyHunters hack
The FBI has confirmed multiple arrests related to the ShinyHunters hack. The arrests represent another apparent takedown linked to the hacking group.
- FBI confirms ‘multiple’ arrests related to ShinyHunters hack — theregister.com
Legacy sign-on service comes back to bite school software provider Bromcom
School software provider Bromcom was hit after intruders exploited a legacy sign-on service that was still running for an internal system. The attackers retrieved email addresses from the superseded technology.
Millions of Airbnb, Uber, and Google records are for sale, hackers claim
A hacker using the name Marx claims to be selling 54 million records tied to Airbnb, Uber, PayPal, Booking.com, and Google. Cybernews researchers say the samples appear to come from one third-party SMS provider rather than separate company breaches, and include phone numbers, mobile carriers, and some Uber rider names. If genuine, the data could enable phishing, impersonation, and possible account takeover attacks.
For your eyes only: WhatsApp tests Restricted Chat feature
WhatsApp is testing a Restricted Chat feature for Android and iOS beta users that keeps selected one-on-one conversations on the primary phone only. When enabled, restricted chats stop delivery to linked devices, prevent automatic media saves, block chat exports, and disable Meta AI use in the chat. The feature is currently only in beta, with no clear timeline for a wider rollout.
- For your eyes only: WhatsApp tests Restricted Chat feature — cybernews.com
FitnessKPI denies data breach but confirms some logs still spilled
FitnessKPI denied a data breach, saying someone accessed its platform using valid customer administrator credentials. Files shared online appear to expose data for 1,848 Genae Écully gym members, including contact details, membership records, bookings, contracts, and payment-related information. Researchers warned criminals could use the records for convincing phishing messages and fake payment reminders.
Most workers have encountered cyber threats at work, EU survey says
A Eurobarometer survey of 25,747 EU citizens found that three in four employees have encountered suspicious emails, messages, or links at work, with many also reporting phishing, malware, and AI-generated scams. While 72% say they can identify phishing emails, only 54% check senders before clicking and fewer than half could recognize deepfake videos. Six in ten workers received cybersecurity training in the past year, but 85% want to improve their skills.
Meta Fixes Muse VM Escape Vulnerability
Meta rushed to fix severe security vulnerabilities in its Muse AI agent shortly before launch, including a KVM escape that could let users break out of an isolated virtual machine and access sensitive internal Meta databases and services; the issues reached Mark Zuckerberg and prompted overtime fixes, with sources citing “half-baked protections” and engineer fears of a major data breach. Separately, an Oklahoma federal judge ruled that a police officer violated the Fourth Amendment by searching a woman’s California license plate in Flock’s automated license plate reader system and using her travel history to justify searching her car.
- Meta Rushed to Fix Muse ‘VM Escape’ Vulnerability Soon Before Launch — 404media.co
- Muse escapes containment — 404media.co
🖥️ Hardware & Devices
Linux 7.4 To Mainline Support For The Google Tensor G5, Pixel 10 Devices
Linux 7.4 is set to mainline support for Google’s Tensor G5 SoC and Pixel 10 devices. The pull request introduces the new ARCH_GOOGLE architecture, Tensor G5 SoC family support, and initial Device Trees for the Pixel 10, Pixel 10 Pro, and Pixel 10 Pro XL. Tensor G5 includes Cortex-X4, Cortex-A725, and Cortex-A520 cores, Imagination PowerVR graphics, LPDDR5X memory, and a 4th Gen TPU.
Linux 7.4 Continues Preparing For Intel’s Rugged Panther Lake
Linux 7.4 is adding further enablement for Intel’s Rugged Panther Lake (Panther Lake R), a ruggedized variant believed to be the Starfire SoC. The patches add support across intel-uncore-freq, Intel PMC, intel_tcc_cooling, and turbostat, mostly by adding the PANTHERLAKE_R device ID. Linux 7.4 is expected to become the baseline for Rugged Panther Lake/Starfire due to power-management requirements.
Qualcomm Patching Linux For Hardware Thermal Binning To Benefit Snapdragon X Laptops
Qualcomm has submitted Linux kernel patches to add hardware thermal binning support, allowing the OS to properly account for different junction temperature limits in otherwise identical Qualcomm SoCs. The support is aimed at benefiting Snapdragon X laptops, including the Snapdragon X1 Elite (X1E80100) and newer chips, by reading manufacturing fuses to identify each silicon’s thermal bin. The patches are currently out for review on the Linux kernel mailing list.
Linux 7.4 AMD P-State Driver Will Be Ready For Ryzen Zen 6 CPUs
The AMD P-State driver changes queued for Linux 7.4 will add initial support for AMD Ryzen Zen 6 client processors, including maximum CPU frequency capacity handling and updated Energy Performance Preference tuning for standard, dense, and new low-power cores. The pull request also includes several bug fixes. Linux 7.4 stable is expected around the end of December, positioning it as a baseline for Zen 6 client CPUs expected in early 2027.
The Pulse: a new trend of CPU shortages
A new CPU shortage is affecting cloud customers, with cheap spot pricing largely disappearing and reservations requiring months of advance planning. Demand from AI workloads like reinforcement learning and general-purpose agents is straining supply, making it difficult even for large companies to secure enough CPU capacity. The article warns that the situation may worsen before it improves.
- The Pulse: a new trend of CPU shortages — blog.pragmaticengineer.com
Ghost emerges from stealth with $11M seed led by a16z for $3,499 AI agent PC with RTX Pro 4000 GPU
Ghost, founded by 19-year-old Zain Javaid, emerged from stealth with an $11 million seed round led by Andreessen Horowitz. Its first product, Core, is a $3,499 personal computer designed specifically to run AI agents, featuring an Nvidia RTX Pro 4000 SFF Blackwell GPU and preinstalled models. Preorders opened Monday, with shipping scheduled for the last week of October.
- Ghost emerges from stealth with $11M seed led by a16z for $3,499 AI agent PC with RTX Pro 4000 GPU — techcrunch.com
Germany’s RobCo hits $1B robotics unicorn valuation
Munich-based robotics startup RobCo has reached a valuation exceeding $1 billion through a $40 million employee secondary share sale, doubling its roughly $500 million valuation from January. Existing investors Sequoia and Lightspeed participated alongside new backers Cherry Ventures and European Tech Collective, and RobCo plans to commercially launch its two-armed, self-learning Alfie industrial robot in 2027 while expanding in the US manufacturing market.
- Europe’s new robotics unicorn: Germany’s RobCo hits $1B valuation — techfundingnews.com
- Munich’s RobCo sells $40M shares at $1B valuation, up from $500M — wsj.com
Greenvolt begins building 600 MW/2.4 GWh BESS in Poland
Greenvolt Power has started construction of a 600 MW/2.4 GWh BESS in Siedlce, Poland, with BYD Energy Storage supplying the battery technology and commercial operations expected by the end of 2027. The project will connect to PSE’s 400 kV transmission network and charge during high renewable generation, discharging when demand rises. Once it operates alongside Greenvolt’s Turośń Kościelna and Ełk projects, the company will have 1 GW/4 GWh of battery storage capacity in Poland.
- Greenvolt begins building 600 MW/2.4 GWh BESS in Poland — ess-news.com
The designer lamps in my house
The author tours his collection of designer lamps, including the Artemide Tolomeo Mini, Richard Sapper’s Tizio Micro, and Isamu Noguchi Akari lamps, detailing their design histories, materials, and personal use. He says they are expensive but worth it to him as collectibles rather than just practical lighting, and notes that similar well-designed lamps can be found more cheaply at IKEA or vintage sources.
- The designer lamps in my house — arslan.io
Huawei and Qualcomm Announce Broad Patent License Agreement
Huawei and Qualcomm announced a multi-year, broad patent license agreement that includes cross-licenses covering 5G, compute, AI, and networking, along with Qualcomm’s purchase of certain Huawei U.S. patents in compute, AI, networking, and other areas. The transaction will close after necessary regulatory approvals and reflects both companies’ commitment to intellectual property rights and FRAND licensing principles.
A tribute to one of the best games on the Atari 2600
This tribute recreates Carol Shaw’s Atari 2600 classic River Raid by rebuilding the original ROM one 6502 instruction at a time in JavaScript and verifying it frame-by-frame against a real ROM. It preserves the endless river, fuel and scoring mechanics, enemy behaviors, crashes, and hardware-accurate collision handling. The page also acts as a playable guide, detailing controls, fuel depots, lives, sections, and point values for tanks, helicopters, jets, bridges, and depots.
- A tribute to one of the best games on the Atari 2600 — plicerin.github.io
Windows 11 26H2 blighted by new audio issue
A new audio issue is affecting Windows 11 26H2 and involves an AC-3 decoder problem. The same AC-3 decoder issue also impacts Windows 11 25H2 and 24H2.
- Windows 11 26H2 blighted by new audio issue — theregister.com
The question you would ask HPE and NVIDIA if nobody was recording
The Register is hosting an off-the-record data sovereignty dinner in Washington on 29 October, bringing together AI infrastructure leads from HPE and NVIDIA. The event is framed around what candid questions attendees would ask if no one were recording.
- The question you would ask HPE and NVIDIA if nobody was recording — theregister.com
What people say about their file infrastructure once the badge comes off
Joe Fay will chair a Register dinner in Soho on 20 October. Infrastructure leaders will compare notes off the record on the file estates they inherited. The event promises candid discussion once the badge comes off.
- What people say about their file infrastructure once the badge comes off — theregister.com
Xbox says this Samsung smart TV can no longer be used to play games
Microsoft will end Xbox Cloud Gaming support for 2020 Samsung smart TVs on November 3, saying the older models cannot receive the software updates required by the service. Affected users must switch to a 2021 or newer Samsung TV or use another device, such as a smartphone, tablet, or Xbox console, to keep playing.
🌍 Tech & Society
What are you doing this week?
The article asks readers what they are doing this week and invites them to share their plans. It also notes that it is acceptable to do nothing at all.
Becoming a photographer
The article reflects on AI’s impact on programming after DHH’s Rails World keynote, arguing that programmers must adapt like painters becoming photographers. The author describes building HEY CLI and designing a calendar TUI through AI-assisted “vibe coding,” with almost no human code review. While roles blur, deep expertise remains valuable for guiding AI and expanding what can be imagined.
- Becoming a photographer — stanko.io
There was a light that never went out
An article discusses a GCP incident report about networking failures in the us-central1 region, where a technician accidentally disconnected fiber-optic cables during routine maintenance. The technician installed higher-density optical transceivers incompatible with those still in the network fabric, causing loss of connectivity despite redundant routers and a planned one-at-a-time upgrade. The author uses the incident to argue that networking systems are inherently unsafe and need error-tolerant designs with cross-checks and recovery mechanisms.
- There was a light that never went out — surfingcomplexity.blog
SpaceX closes up 7.6% on Morgan Stanley ‘cheap’ call, highest since mid-June, Musk a trillionaire
SpaceX shares rose 7.6% Monday after Morgan Stanley called SPCX “cheap” and set a $300 price target, reaching their highest level since mid-June and returning Elon Musk to trillionaire status. The rally followed SpaceX’s record IPO and recent launches, with analysts bullish ahead of the next Starship test flight and third-quarter earnings.
- SpaceX closes up 7.6% on Morgan Stanley ‘cheap’ call, highest since mid-June, Musk a trillionaire — cnbc.com
Anthropic, Google, OpenAI, Meta, ex-Anthropic researcher testify at NYC Council AI safety hearing
Former Anthropic researcher Jacob Coxon testified at a New York City Council hearing on AI safety, warning that the race to develop AI poses risks the industry does not fully understand. Coxon said, “We do not know how to control any AI system yet,” as lawmakers weighed imposing safeguards on the global AI industry. Representatives from Anthropic, Google, OpenAI, and Meta also appeared at the hearing.
- Anthropic, Google, OpenAI, Meta, ex-Anthropic researcher testify at NYC Council AI safety hearing — bloomberg.com
CFTC proposes US framework for crypto exchanges’ retail leveraged/margined spot trading, no Congress
The CFTC proposed a federal regulatory framework allowing crypto exchanges to offer retail customers leveraged and margined spot trading, including a new “crypto asset market” registration category via Regulations CTX and CAM. The agency said it can proceed without congressional action but cannot require crypto assets to trade on CFTC-registered platforms absent Congress.
- CFTC proposes US framework for crypto exchanges’ retail leveraged/margined spot trading, no Congress — theblock.co
OpenAI to watermark ChatGPT under EU rules
OpenAI outlined a phased plan to comply with EU AI Act text provenance rules, starting with global API customers able to opt in to watermarking for select models (off by default), while eligible ChatGPT and Codex text output in the EU will receive an invisible watermark in coming weeks. OpenAI will also grant approved researchers and expert organizations limited access to its text watermark detector, noting textGrain’s limitations on short, constrained, or edited text and that its image and audio provenance tools remain publicly available.
- OpenAI to add EU AI Act-required text watermarking for ChatGPT/Codex; opt-in API globally — openai.com
First US pilot: Nolla Health uses AI to diagnose, prescribe acne meds in Utah, no human oversight
Nolla Health will launch a first-of-its-kind pilot in Utah using AI to diagnose acne and issue prescriptions without direct human oversight. The program, limited to acne medications, marks the first time a state has allowed AI to replace a doctor’s role in such care. It follows months of scrutiny and anger from doctors over Utah’s efforts to expand AI in healthcare.
- First US pilot: Nolla Health uses AI to diagnose, prescribe acne meds in Utah, no human oversight — bloomberg.com
Q&A with AI researchers on leaving labs, AGI risks, IPOs, safety, oversight (NY Mag)
New York Magazine, in collaboration with Asterisk, interviews eight former employees from OpenAI, Anthropic, and Google DeepMind about why they left their AI labs. The Q&A covers concerns over AGI dangers, defense partnerships, corporate drift, AI safety and oversight, planned IPOs, and debates over whether to quit in protest or push for change from within.
Nico Caprez, Huang’s son-in-law, now Nvidia VP, his ‘consigliere’ and main neoclouds contact
Nico Caprez, Jensen Huang’s son-in-law, has rapidly risen to become an Nvidia VP and one of the CEO’s most trusted executives, serving as his “consigliere” and main contact for neoclouds. His elevated role comes shortly after his marriage to Madison Huang, a rising Nvidia employee and Jensen Huang’s daughter, who recently celebrated their wedding in Hawaii.
- Nico Caprez, Huang’s son-in-law, now Nvidia VP, his ‘consigliere’ and main neoclouds contact — theinformation.com
Businesses Struggle to Budget AI Spending
A survey of nearly 400 businesses found that only 11% could accurately forecast their AI spending, as companies that encouraged heavy AI use now face harder-to-track costs. A separate study of more than 6,800 tasks found that lower-priced AI models cost more than higher-priced ones on 32% of tasks, largely because token usage is unpredictable.
- Survey: Only 11% of 400 firms can forecast AI spend; cheaper models cost more on 32% of 6,800+ tasks — wsj.com
Billionaires Index: Tech billionaires gained all $845B in 2026; non-tech lost $62B
Tech billionaires have accounted for all $845 billion in wealth gains so far in 2026, while billionaires whose fortunes came from outside tech lost a combined $62 billion, according to Bloomberg’s Billionaires Index. The AI-driven valuation boom has created unprecedented wealth, but the article suggests the peak may have just passed.
Ex-Groq engineers sue in Delaware over Nvidia $20B acqui-hire, alleging employees were short-changed
Former Groq engineers have sued Groq in Delaware over its $20 billion “non-exclusive” acqui-hire deal with Nvidia in 2025. The plaintiffs allege the transaction short-changed employees by leaving them out of the deal.
- Ex-Groq engineers sue in Delaware over Nvidia $20B acqui-hire, alleging employees were short-changed — ft.com
OpenAI Tests Visual Ads in ChatGPT
OpenAI will begin testing a new visual ad format in ChatGPT in the U.S. later this month, showing clearly labeled ads during image generation with an initial group of advertisers. The ads will remain separate from generated images and will not influence ChatGPT’s answers. The test is initially limited to the U.S. as OpenAI expands advertising and adds measurement and brand-suitability tools.
- OpenAI to flood your eyeballs with visual ads — theregister.com
- OpenAI to test labeled visual ads in ChatGPT in US this month during image generation — bleepingcomputer.com
Alpha: invite-only Discord of ~30 Gen Z Polymarket traders pooling research/capital, $40M since 2024
Alpha is an invite-only Discord server of about 30 mostly Gen Z Polymarket traders who pool research and capital. Since 2024, the group has reportedly made an estimated $40 million by treating prediction markets around elections, wars, and disasters as “free money on the floor.”
- Alpha: invite-only Discord of ~30 Gen Z Polymarket traders pooling research/capital, $40M since 2024 — ft.com
OKX-ICE JV seeks SEC nod for tokenized shares of 63 NYSE firms under new US rules
OKX filed with the SEC on Sunday to launch a tokenized-stock trading platform, seeking approval through OKXICE, its joint venture with Intercontinental Exchange. The platform plans to offer tokenized shares of 63 NYSE-listed companies under new US rules, with issuers given 30 days to opt out before trading begins.
Firmus to allocate ~50% of IPO to existing holders as demand far exceeds offer in Aussie mega-IPO
Firmus Grid Ltd. plans to allocate about 50% of its IPO shares to existing shareholders. The decision comes as investor demand far exceeds the offer size in one of Australia’s largest IPOs.
- Firmus to allocate ~50% of IPO to existing holders as demand far exceeds offer in Aussie mega-IPO — bloomberg.com
The mental health of young men is declining. Experts warn it could get worse
The mental health of young men has declined since 2012, with self-reported mood disorders among men aged 15–24 rising from 11% to 16% and anxiety disorders from 4% to 10%. Experts warn the trend could worsen, citing social isolation, social media use and online sports betting, while men are three times as likely as women to die by suicide. The federal government launched a consultation this year to investigate why boys and men are suffering.
After bankruptcy, he was banned from sports betting sites. Then he found Kalshi
After filing for bankruptcy and self-excluding from DraftKings and FanDuel, Thomas was drawn into the prediction market Kalshi by an Instagram ad and soon fell more than $25,000 into debt. Kalshi dismissed his case as cherry-picked and defended its exchange model, but gambling counselors say his experience is increasingly common as Kalshi and Polymarket run nationwide ads and avoid state-mandated protections for problem gamblers.
AI Companies Are Parasites
The article argues that AI companies operate like parasites by scraping the internet and books without compensation, training models on that data, and selling access while harming the creators and communities they depend on. It criticizes licensing deals as insufficient and points to local costs such as pollution, higher utility rates, and few lasting jobs from data centers. The author concludes that AI firms are now deeply embedded in society and may be too difficult to dislodge.
- AI Companies Are Parasites — coryd.dev
How did Rosalind Franklin miss the helix in her iconic DNA image? She didn’t
The article challenges the claim that Rosalind Franklin missed the helical structure in her iconic DNA image, arguing she did not miss it. It seeks to correct the misconception that she failed to recognize the helix in her own data.
One person is now a quorum at the SEC
FT Alphaville reports that one person is now a quorum at the SEC. The article is otherwise behind a registration wall, so no further details are available.
Dutch government to ban smartphones from schools
The Dutch government plans to ban smartphones in schools during lessons and breaks to reduce distractions and better protect children online, State Secretary Willemijn Aerdts announced Monday. Aerdts hopes to implement the “At Home or in the Vault” measure on January 1, 2027, while many schools already follow similar policies. The ban is part of a broader package that includes a proposed European minimum age for social media and stricter safety requirements for tech services.
- Dutch government to ban smartphones from schools — nltimes.nl
Norway Plans Temporary Smart Glasses Ban
Norway has proposed a temporary ban on camera-enabled smart glasses in many public places, including parks, beaches, museums, schools, gyms, and public events, while allowing private use. The government cites privacy concerns over people being photographed, filmed, or audio-recorded without their knowledge, though the bill still needs support from other parties to pass. Similar restrictions are being considered in the UK, US, and elsewhere as devices like Meta’s Ray-Ban smart glasses grow in popularity.
- Norway plans temporary ban on smart glasses — theguardian.com
- Norway Eyes Partial Ban of Smart Glasses — barrons.com
Google Accused of Felling 300 Hectares of Forest for a Finnish Data Centre
More than 300 hectares of forest were cleared for a Google data centre in northern Finland without the environmental assessment Finland’s regulator said was required, with satellite images showing felling continued until August despite a May letter. The regulator’s Sari Myllyoja said it appeared Google’s project company proceeded regardless, while Google said the felling complied with the Forest Act. The regulator is investigating possible measures, both sites are under building bans, and Muhos municipality says it issued no building permit.
- Google Accused of Felling 300 Hectares of Forest for a Finnish Data Centre — woodcentral.com.au
The AI boom is making the cheapest smartphones disappear
AI data centers are driving a global memory-chip shortage, causing the world’s cheapest smartphones to disappear as component costs rise. Existing smartphone prices have increased about 15% globally this year and new models are roughly 25% more expensive, prompting Chinese makers to cut entry-level projects and focus on premium devices. The shift threatens to widen the digital divide by making first smartphones less affordable.
- The AI boom is making the cheapest smartphones disappear — restofworld.org
Green party votes to adopt policy that equates Zionism with racism
The Green Party of England and Wales passed a binding motion equating Zionism with racism and calling for Israel to be replaced by a single Palestinian state with equal rights for all. Four of its five MPs opposed the motion, while Jewish members and lawyers warned it could breach the UK Equality Act 2010 by discriminating against Zionists and Jewish institutions. Israel’s foreign ministry called the resolution antisemitic and barred Green leader Zack Polanski from entering Israel.
- Green party votes to adopt policy that equates Zionism with racism — theguardian.com
The future of independence is interdependence
The article argues that independence is really interdependence, using the author’s life in one room with mobility equipment, assistive technology, and help from family to show that visible dependence can enable a functioning, productive life. It challenges the myth of self-sufficiency, noting that everyone relies on infrastructure, services, and other people, often invisibly.
People are asking ChatGPT to help them decide how to vote in the midterms
As midterm voting proceeds, some voters are using AI chatbots like ChatGPT to research ballots, compare candidates, and check viral claims, especially in lower-profile races. Experts warn these tools can be subtly persuasive and may reinforce users’ existing views rather than deepen their understanding, with unknown biases due to opaque training and sourcing.
Altman Urges Accepting AI Risks for Benefits
OpenAI CEO Sam Altman said the world should accept some harmful AI outcomes, such as hacks and scams, in exchange for AI’s benefits and a lighter regulatory approach, arguing such trade-offs are necessary to advance the technology. His comments sparked backlash and followed an OpenAI safety expert’s resignation over a broken safety culture, summer safety incidents involving OpenAI agents, and a White House pact favoring corporate self-policing.
- Accept ‘bad things’ in return for benefits of AI, says Sam Altman — theguardian.com
- Altman: The world should accept some bad things happening for the benefits of AI — politico.com
“I’m Embarrassed on Behalf of the Tech Industry”
Tech worker Ben Thompson said he felt embarrassed on behalf of the tech industry while helping his mom manage her digital life because everything was too hard and convoluted. The author echoed that embarrassment, citing similar complaints from non-tech family and friends that digital products are confusing and difficult to use.
- “I’m Embarrassed on Behalf of the Tech Industry” — blog.jim-nielsen.com
Hacker News: An Apology
Lee Phillips withdrew his recommendation of Hacker News after a 1,101-comment thread became a toxic political flamewar filled with insults, obscenity, and racism, with dissenting comments marked dead while moderators appeared absent. He emailed the moderators to ask whether the site’s guidelines had changed, received no reply, and later saw only limited moderation, including a warning to a respectful commenter.
- Hacker News: An Apology — lee-phillips.org
Observing Earth’s Shadow from a Plane
During a dawn flight, the author photographed Earth’s shadow cast on the atmosphere and built a simple spherical-Earth model to predict the dark band’s location in each of 34 photos. The model derives the shadow geometry and uses color changes to locate the band’s edge, while ignoring effects such as refraction, the Sun’s disk, extinction, and multiple scattering so its predictions can be compared with the photo data.
- Observing Earth’s Shadow from a Plane — hermandaniel.com
Steve Jobs died 15 years ago today. I built an interactive tribute
Akash Bangad has created an interactive tribute marking 15 years since Steve Jobs died on October 5, 2011. The experience lets users explore screens of machines Steve Jobs made, hear his voice, and light a candle, and it references Bangad’s film Steve Jobs’ Last Idea (He Predicted AI in 1983).
- Steve Jobs died 15 years ago today. I built an interactive tribute — thankyousteve.org
Nobel Prize in Physiology or Medicine 2026
The article outlines the background of the Nobel Prize in Physiology or Medicine, noting it was the third prize area mentioned in Alfred Nobel’s will and that Nobel had an active interest in medical research, including contact with Swedish physiologist Jöns Johansson around 1890. The prize is awarded by the Nobel Assembly at Karolinska Institutet in Stockholm, Sweden.
- Nobel Prize in Physiology or Medicine 2026 — nobelprize.org
2026 Nobel Medicine Prize winners
The 2026 Nobel Prize in Physiology or Medicine was awarded jointly to Karl Deisseroth, Peter Hegemann, and Georg Nagel for discoveries concerning light-gated ion channels and optogenetics. Hegemann and Nagel discovered the algal protein channelrhodopsin, which Deisseroth turned into a light-controlled switch for nerve cells, enabling researchers to study how nerve cells shape memories, feelings, and behavior in the living brain.
- Press Release: Nobel Prize in Physiology or Medicine 2026 — nobelprize.org
- 2026 Nobel Prize in Physiology or Medicine: Deisseroth, Hegemann, Nagel — nobelprize.org
Girl, 12, fatally shot in US during ‘dispute over parking space’
A 12-year-old girl was fatally shot in Baltimore over the weekend when she was caught in the crossfire of an apparent dispute between two men over a parking space. A woman was also wounded, and police are searching for at least two suspects; no arrests have been made.
US removes all bombers from RAF Fairford base
The US has withdrawn all bombers from RAF Fairford, with a Pentagon official confirming about a dozen B-1s returned to their home stations in America. The move follows a counter-terror investigation into suspicious vehicles near the UK base on 27 September and multiple arrests, while US media reported a new Iran-backed threat prompted the withdrawal.
Sales of sub-€25,000 electric car models set to rise sevenfold
According to T&E’s latest EV progress report, EU CO2 targets are driving a rapid expansion of affordable electric cars, with sales of models starting below €25,000 expected to rise sevenfold in 2026 compared with 2024. Nearly 40 new electric models launched in the first half of 2026, while EU battery-electric vehicle sales reached 1.64 million from January to August, up 45% year-on-year. T&E warns that weakening the 2030–35 targets would halt the affordable-EV ramp-up and harm Europe’s competitiveness in the global EV race.
- Sales of sub-€25,000 electric car models set to rise sevenfold — transportenvironment.org
Hacker News addiction and taking simple mundane breaks in life
The author reflects on their Hacker News addiction, arguing that using the site as a break often turns into a lengthy distraction that leaves them in motion but not going anywhere. They suggest taking simple, mundane breaks and note that ambition can be pursued through everyday life rather than as a grand end in itself. The post was written while procrastinating before an exam and is framed as a personal reflection meant to help others with similar habits.
- Hacker News addiction and taking simple mundane breaks in life — smileplease.mataroa.blog
Replacement of petroleum based products with plant-based materials (2025)
The article titled “Replacement of petroleum based products with plant-based materials (2025)” could not be accessed because its content requires JavaScript and cookies to be enabled. No substantive details from the article are available beyond its title.
- Replacement of petroleum based products with plant-based materials (2025) — onlinelibrary.wiley.com
Powerless F1 drivers frustrated by Bahrain F1 software glitch
Formula 1 drivers were left powerless and frustrated by an unprecedented power-unit software glitch during the formation laps of the Bahrain Grand Prix, which left multiple cars stuck in idle mode. The FIA rapidly issued a software patch during a red-flag stoppage, causing a 50-minute delay before the race could start. Drivers including Lando Norris and Sergio Perez called the incident horrible, unacceptable, and further evidence of problems with F1’s complex hybrid electronics.
- Powerless F1 drivers frustrated by Bahrain F1 software glitch — motorsport.com
The 15-year search for a band that charted once and vanished
A 15-year search for the band Salvage, whose professionally produced tracks “Electric” and “Searchlights” appeared on a CD but otherwise seemed to vanish, led to repeated dead ends. Using AI-assisted research, the author confirmed a single Official Charts entry: No. 47 on the Independent Singles Chart for 4–10 June 2006, linked to the obscure label Despatches, but further investigation remained unresolved.
- The 15-year search for a band that charted once and vanished — shahidhussain.com
1 in 8 cancer cases are caused by infections
A new study estimates that about 1 in 8 cancer cases—roughly 2.3 million in 2024—are caused by viruses, bacteria, or parasites. Four pathogens account for most infection-related cancers: Helicobacter pylori, HPV, hepatitis B, and Epstein-Barr virus. Experts say preventing these infections through tools such as vaccines is key to reducing cancer rates, though HPV vaccination coverage remains well below global targets.
- 1 in 8 cancer cases are caused by infections — livescience.com
Windows 11 market share creeps upwards as cost of staying on Windows 10 set to double
Windows 11 market share is continuing to rise as the cost of keeping Windows 10 devices on security updates is set to double. The increase raises the price for users and organizations that delay upgrading from Windows 10.
- Windows 11 market share creeps upwards as cost of staying on Windows 10 set to double — theregister.com
BT buys TalkTalk out of administration to keep 2.5M customers connected
BT has bought TalkTalk out of administration to keep around 2.5 million customers connected. The rescue carries an estimated £400 million cash impact, with ministers intervening and a competition review looming.
- BT buys TalkTalk out of administration to keep 2.5M customers connected — theregister.com
Buyer missed one option and got 1% of the cable they needed
A buyer overlooked an option and received only 1 percent of the cable they needed. The supplier, described as silly, ended up short-changing itself in the process.
- Buyer missed one option and got 1% of the cable they needed — theregister.com
Iran claims it knocked Starlink offline, warns satellites are now military targets
Iran’s cyber chief claimed Tehran deliberately disrupted Starlink during January’s nationwide internet blackout amid anti-regime protests, saying Iran was the first country to successfully shut down the service. He also warned Starlink could become a “legitimate military target” if used for warfare, while Tehran renewed a diplomatic dispute with Norway over claims the service is used for espionage.
US arrests tech executive over alleged $300M GPU diversion to China
US authorities arrested Greg Lui, a technology company owner, over an alleged $300 million scheme to divert US-made servers containing restricted GPUs to China. Prosecutors say he used false documents and routed the shipments through Malaysia and Singapore. Lui faces charges including export control violations, smuggling, and money laundering, with potential decades in prison if convicted.