Alvaro Lopez Ortega / 2026-10-28 Briefing

Created Fri, 30 Oct 2026 06:05:37 +0000 Modified Fri, 30 Oct 2026 06:06:55 +0000
11225 Words

Suspected China-linked hackers exploited PHP2Shell and a Linux flaw to breach a US government-affiliated network, compromising 24 firewalls and stealing Active Directory credentials. Russia approved guidelines to camouflage data centers with anti-drone nets and underground sites amid Ukrainian attacks. In South Korea, a voice-phishing gang intercepted a victim’s calls to police and regulators, stealing about $21,000 over three days.

🔐 Security & Privacy

China-linked hackers exploit PHP2Shell to breach US government network

Suspected China-linked hackers exploited PHP2Shell and a Linux flaw called Fragnesia to breach a US government-affiliated network, compromising 24 firewalls after scanning 19,000 internet-facing devices across four countries. They stole Active Directory credentials and attempted to create hidden accounts, though some attempts remain unconfirmed.

South Korean voice phishing gang intercepted victim’s calls to police and regulator

A South Korean voice-phishing gang impersonating prosecutors convinced a man he was under investigation, isolated him in a motel, and used a malicious remote-control app to intercept his calls to the Financial Supervisory Service and the 112 emergency hotline. Over three days, the scammers forced him to write 45 self-reflection letters and stole about 31 million won ($21,000), while trying to extract another 42 million won ($29,000). Police eventually reached the victim by playing radio transmissions, breaking the gang’s psychological control.

Security vulnerability at Bitonic allows unauthorized access to customer accounts

Bitonic suffered a security incident after attackers exploited a PDF-processing library vulnerability to access customer accounts, withdrawing bitcoin from three accounts and changing login credentials for two. The company reimbursed affected customers, says the flaw has been fixed, and is investigating whether other customers’ personal data was viewed. Bitonic is advising customers to switch from email/SMS two-factor authentication to app-based one-time passwords.

9-Bit AES Key Case

Several PDF implementations that advertise AES-256 generate file keys from small or insecure random state spaces, enabling passwordless decryption. MuPDF’s AES-256 key depends on only 9 bits of random state, while Foxit uses time-based randomness, making keys trivial to recover.

GitLab CVE Exploitation Attempt

An attacker came close to exploiting critical GitLab vulnerability CVE-2026-97229 on an Alpine GitLab instance on October 20–21, though no compromise was confirmed; the instance was treated as compromised and rebuilt from scratch. Checks found no tampering with the aports repository, packages, or October 21 releases, and all passwords and GitLab secret keys were reset. GitLab remains closed for testing before reopening, and users are advised to change any reused passwords.

Fingerprinting Script Bioweapons Mystery

A researcher crawling 10,000 websites for canvas fingerprinting scripts found an obfuscated ServicePipe script containing alarming function names such as “computeRicinLethalDosageForWaterSupply.” The author suspects these names are intentionally misleading anti-LLM obfuscation rather than evidence that the script creates bioweapons.

Google unveils Google Multi-Party Compute (MPC), a digital-asset custody and security service in …

Google has unveiled Google Multi-Party Compute (MPC), a private-preview digital-asset custody and security service that splits crypto private keys across separate environments and enforces transaction rules before signing. Several major financial institutions are testing it, and Alphabet is using it to secure its own digital assets on Ethereum and Solana. Google plans commercial availability in early 2027 and says the system reduces single-machine compromise risk and can eventually support post-quantum cryptography.

US grand jury indicts Saudi OpenAI engineer Abdul Alfozan on felony interception charge

A federal grand jury indicted Abdulrahman “Abdul” Alfozan, a 35-year-old OpenAI engineer and Saudi national, on a felony charge of interception of oral communications. Prosecutors allege he installed an encrypted listening device behind a television in an OpenAI headquarters conference room used for confidential discussions; he was arrested Oct. 9 and remains in custody without bail.

Upwind Security raises $400M led by Bessemer at $3.8B valuation

Upwind Security raised $400 million in a funding round led by Bessemer at a $3.8 billion valuation, up from $1.5 billion in January. The AI and cloud cybersecurity startup has now raised $830 million in total funding.

Homeland Security Employee Data Leak

A leak published by Distributed Denial of Secrets as “Icepick” exposed contact information and other records for more than 40,000 DHS employees and contractors across components including ICE, CBP, and HSI, with references to internal systems and undercover and surveillance programs. The group says the data came from DHS’s Okta single sign-on system; Okta says it has no evidence its systems were compromised, while DHS says it is investigating. DDoS said it vetted and cleaned the data to map organizational structures and identify officials involved in DHS policy and operations rather than serve as an indiscriminate employee directory.

Scanning 7.25M Public Repositories on China’s Largest Git Platform for Secrets

Security researcher Ben Zimmermann scanned 7,256,531 public repositories on Gitee, China’s largest indigenous Git hosting platform, using TruffleHog and Kingfisher. The scan found 26,400 unique verified-live credentials, including an active organization-admin credential for a major enterprise AI platform and thousands of live cloud provider access keys.

Newly discovered Kerberos delegation abuse path

Researchers described a new Kerberos delegation abuse path that combines Reflective RBCD, U2U sacrificial-user techniques, and an Impacket extension for S4U2Proxy with U2U to obtain a forwardable ticket for Constrained Delegation without Protocol Transition when the delegation-trusted account has no SPN. The technique requires RC4 to be permitted, destroys the affected user account, and needs MAQ enabled in the domain or an already-compromised object with an SPN.

ICE Is Running One of the Largest Anti-Left Surveillance Ops in US History

Attorneys for 15 Minnesota residents facing federal conspiracy charges have obtained evidence that ICE/DHS conducted a sweeping surveillance operation against left-wing and anti-Trump immigration-policy activists, using undercover agents, drones, Patriot Act tools, and church infiltration under programs tied to NSPM 7. Defense attorneys say the Minnesota 15 prosecution is part of a broader nationwide crackdown on administration opponents and compare it to COINTELPRO; three former law enforcement officials who reviewed filings concur. The charges stem from activists’ efforts to monitor DHS agents during Operation Metro Surge, which prosecutors call a criminal conspiracy but defense lawyers say does not match the alleged conduct.

TLP Is Not a Suggestion

Akamai Security Intelligence Group is described as a global team of researchers, engineers, strategists, and data scientists spanning multiple security disciplines. It draws on sources including Akamai Cloud, open sources, third-party collaboration, and dark web intelligence, and uses its own algorithms and tools to produce research and keep Akamai security solutions current.

Google can now use your Gmail data for ads: How to turn it off

Google changed its privacy policy on October 16 to allow data from connected Gmail, Drive, and Photos apps to inform ads in Search through its Personal Intelligence features. The change applies only if users opt in, and users can disconnect those apps or disable personalization in Google Account settings. However, disconnecting does not erase data already saved, which may still be used for product improvement and AI training.

Oracle celebrates end of Cybersecurity Awareness Month with quadruple trouble

Oracle is facing quadruple security trouble at the end of Cybersecurity Awareness Month. Attackers are probing another EBS bug, while ShinyHunters’ PeopleSoft 0-day has been found to involve three vulnerabilities.

Serial ransomware crook gets 70 months behind bars for Nefilim, Hive roles

Artem Stryzhak has been sentenced to 70 months in prison for his roles in the Nefilim and Hive ransomware gangs. He joined both groups, which targeted businesses and a hospital during the COVID-19 pandemic.

Cisco open sources tool for detecting AI agents on hacking sprees

Cisco has open-sourced Project Quarry, a tool designed to detect AI agents conducting hacking sprees. The project aims to provide a shared language for evaluating agent behavior.

123 Reg discloses ‘AI cyberattack’ on customer data, suspects China nexus

123 Reg has disclosed an “AI cyberattack” on customer data and suspects a China nexus. Company chiefs believe the attackers may be the same group that used ARTEX to breach South Korea’s financial giants.

Objecting to Meta’s AI Training

Meta updated its commercial terms to let it use content from business accounts and pages to develop and train AI, prompting backlash from artists and small businesses. After criticism, Meta delayed the effective date to November 20 and narrowed some language, though the terms still give it broad data-use leeway. Users in the UK and EU can object through Instagram’s privacy center by submitting a privacy rights request, with some users reporting approval under GDPR and the EU AI Act.

Victory! You Can Now Exclude Apps from Siri AI

Apple’s iOS 27.2 introduces an “Excluded Apps” setting that lets users block Siri AI from accessing selected apps’ contents, fulfilling a key request from the Encrypt It Already campaign. The setting is clearer than earlier controls, though excluding apps disables related Siri AI features such as reading messages or onscreen awareness; Android’s Gemini controls remain less straightforward.

Court Should Curb Abuse of Grand Jury Subpoenas Targeting Social Media Users

The ACLU, EFF, and other civil liberties and trade groups filed a brief urging a federal appeals court in Washington, D.C., to curb government use of grand jury subpoenas to unmask anonymous social media critics. The sealed case involves DHS seeking information about an anonymous Reddit user who commented on an immigration officer in Minnesota, first via an administrative subpoena and then a grand jury subpoena. The groups argue such subpoenas should receive First Amendment scrutiny to prevent intimidation and chilling of protected speech.

💾 Fighting DOGE’s Data Grab | EFFector 38.19

EFF warns that the federal government has been pooling Americans’ private data on an unprecedented scale and using it for purposes they never agreed to. In its latest EFFector newsletter, EFF highlights a recent court victory in its lawsuit over the sharing of federal workers’ confidential personnel files with DOGE. The issue also covers a free tracker blocker coming to Safari, a California net neutrality call to action, and EFFector’s availability as a podcast.

🤖 AI & Machine Learning

Microsoft, Epic Hit by AI Hypocrisy Backlash

Microsoft and Epic Games are facing backlash for issuing DMCA takedowns against browser-based fan ports and clones of games like Minecraft, Call of Duty, Halo, and Fortnite while publicly embracing generative AI, prompting accusations of hypocrisy. The takedowns come amid a surge in AI-powered game decompilations and ports and have also hit indie games such as Slay the Spire 2, Celeste, and RimWorld that contain no AI content and were open-sourced against their developers’ wishes. Critics say the episode highlights unresolved questions over copyright, fair use, and ownership of AI-assisted creativity.

We think we’ve seen tech revolutions before. This expert says we’re dead wrong

Oxford scientist Brianna Rosen warns that comparing AI to past tech revolutions is misleading because AI’s speed, scale, and impact are unprecedented, and experts still disagree on whether AI systems can reliably follow human intent. She says AI errors in intelligence or military settings could escalate into conflict, and argues that governments, companies, journalists, and the public need stronger roles in AI oversight.

Can Claude Code really detect hidden cameras in hotel rooms?

Viral posts claim Claude Code can detect hidden hotel cameras or hack door locks, but experts are skeptical of those claims. The scans rely on real security tools, while AI mainly explains results and connects steps, and hidden cameras that record locally or use mobile connections may not appear in Wi-Fi or Bluetooth scans. Experts say physical inspections of mirrors, vents, chargers, TVs, and other objects facing the bed remain essential.

Meta quietly launched Waddle World, a Tamagotchi-like, AI pet-rearing game, in the Philippines in…

Meta has quietly launched Waddle World, a Tamagotchi-like AI virtual pet app currently available only in the Philippines. Users hatch and care for a pet that responds to journal entries with AI-generated replies and doodles, and they can play mini-games to earn food and toys. The limited release may be a test before a wider rollout.

Chinese AI chipmaker Moore Threads reports Q3 revenue up 1,774% YoY to ~$240M and a ~$7.4M net pr…

Moore Threads reported Q3 revenue of about $240 million, up 1,774% year over year. The Chinese AI chipmaker also posted a net profit of roughly $7.4 million.

Sources: Anthropic’s seven cofounders remain extremely influential and personally tight; Daniela …

The provided content is not a news article but an FT.com security verification error page showing error code CG000 / 403 and request ID a51aa264fc5dcfef. As a result, there are no substantive details about Anthropic’s cofounders to summarize.

Anthropic’s $131.7B in cash/ST investments tops all non-bank US firms except Berkshire, Alphabet

Anthropic disclosed in an S-1 filing that it held $112.4 billion in cash and $19.3 billion in short-term investments as of June 30—$131.7 billion total, surpassing every non-bank U.S. company except Berkshire Hathaway and Alphabet. The company also faces more than half a trillion dollars in future commitments, including $321.4 billion in estimated lease obligations and $265.6 billion in fixed or minimum-guaranteed commitments.

Google AI Won’t Name 2020 Winner

Google’s AI-assisted search and AI Mode drew criticism for refusing to clearly state that Joe Biden won the 2020 U.S. election, instead directing users to official FEC results, even though Google Gemini gave the correct answer. The issue appeared limited to U.S. users and was reportedly fixed for some but not all, raising concerns about AI reliability. Similar evasive answers appeared for some other election cycles, including 2024 and 2012, while the tool answered directly for most presidential elections back to 2000 and varied by user and question.

Particle-1

Particle launched Particle-1, a model post-trained on DeepSeek V4.1 Flash to improve token efficiency and lower the cost of capable AI. Across software engineering benchmarks, Particle-1 uses 28.4%–52.2% fewer output tokens and 34.4%–51.6% less cost per task than the base model while maintaining comparable accuracy. This first release focuses on software engineering, with future releases planned for other domains.

Gemini 4 Carbon

Gemini 4 Carbon is the newly announced next frontier model, described as its strongest yet with state-of-the-art cybersecurity and long-horizon capabilities. It ranks #1 on Artificial Analysis Intelligence Index v5.0 at 59 and leads cybersecurity benchmarks SEC-bench Pro (88.2%), ExploitGym (54.8%), and CWE-bench (71.0%), while showing gains in coding and agentic tasks. It will first be available to select cyber defenders and trusted partners via the Fairwind Program during U.S. pre-release testing, priced at $2 per million input tokens and $10 per million output tokens, with cached input 95% off.

Dayjob: Tax – evaluating AI agents on 50 professional tax assignments

Surge AI introduced DAYJOB: Tax, a benchmark of 50 professional tax assignments across eight practice areas that tests AI agents on tax knowledge, calculations, and professional judgment. In one example, an AI agent correctly evaluated sale dates but missed a consequential residence-requirement deadline; on the leaderboard, Claude Sonnet 5.5 leads with a 10.8% pass rate at a 95% rubric threshold, followed by Claude Fable 5.1 at 9.6% and Claude Opus 5.5 at 9.2%.

Nvidia Nemotron 3.5 Transcribe: open ASR for 40 languages

NVIDIA has released Nemotron 3.5 Transcribe, an open-weight, promptable ASR model that combines transcription across 40 languages, translation for 31 languages to or from English, speaker attribution for 13 languages, and audio question answering, handling recordings up to an hour and eight speakers. It reports top leaderboard results and supports quantized local and data-center GPU deployment. NVIDIA also released the open Granary v2 data pipeline and fine-tuning recipes, with a medical case study showing a 45.8% relative word-error-rate reduction after fine-tuning.

Vibe game development October inflection point

October 2026 marks a new inflection point for “vibe-coded” games, driven by Claude Opus 5.5’s improvements in 3D modeling, music generation, and especially game design. The author built four enjoyable HTML/JavaScript browser games with little guidance and no code inspection, while noting that GPT-6 Astra can build games but lacks Opus 5.5’s sense of fun.

Decompiling Games with AI: Virtueel Zoetermeer

The article discusses using AI to accelerate byte-matching game decompilation, noting that an attempt to decompile Age of Mythology: Gold Edition stalled around 60%. It then describes using Claude to decompile the Dutch educational SimCity-style game Virtueel Zoetermeer from archive.org and fix crashes, producing a stable web-playable port. Virtueel Zoetermeer was originally commissioned by the Zoetermeer municipality in the early 2000s and used in schools to teach urban planning.

Sleep framework: A new way for LLMs to retain knowledge in continual learning

Google researchers propose a “sleep” framework for continual learning in LLMs that adds offline memory consolidation phases to help models integrate new facts and improve reasoning without catastrophic forgetting or fixed knowledge cutoffs. The approach alternates active interaction with sleep stages that consolidate short-term context into long-term parameters using pre-allocated Mixture-of-Experts memory blocks and self-generated practice examples. It reportedly enables retention of long-context knowledge up to 10 million tokens while keeping the model architecture unchanged.

Opus 5.5 causing an identity crisis is real

The author repeated Piotr Migdał’s one-shot experiment using Claude Code and Opus 5.5 to generate a three.js visualization of Italo Calvino’s Invisible Cities, comparing medium-effort and max-effort/ultracode runs. The max-effort run cost about 17.7× more, used far more tokens and API time, and added more code, while the medium-effort run took longer in active wall time. Both results were impressive, though Piotr’s version was slightly better visually and the max-effort version never fixed mouse-input errors; the author shared the interactive versions, source code, and transcripts and encouraged others to try the prompt.

Artificial Analysis Intelligence Index v5.0

The Artificial Analysis Intelligence Index v5.0 combines nine evaluation datasets to assess language models across knowledge work, coding, science, and general capabilities, with each category weighted 25%. It includes tasks such as agentic file outputs, SaaS automation, long-PDF grounding, code changes, terminal execution, and open-answer accuracy/hallucination scoring. The methodology emphasizes fairness and real-world applicability while noting that the index, like all evaluation metrics, has limitations.

Artificial Analysis Intelligence Index v5

Artificial Analysis released Intelligence Index v5 on October 27, 2026, adding two new benchmarks and three upgrades, with 60% of the index now based on private test sets or held-out answers; it also launched Progress Index v1. Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55), while the open-weights gap widened from 3 to 12 points. Among leading models, cost per task varies more than 13x, from $0.96 for GPT-6.1 Terra (max) to $13.04 for Claude Fable 5.1 (max).

Please state the nature of the medical emergency: Google is building a robot medic for NASA

Google is developing an AI medical assistant for NASA that is still undergoing testing. NASA now reportedly wants to give the system hands, moving it toward becoming a robot medic.

Amazon tells tokenmaxxing devs to slow their AI roll

Amazon is urging developers to slow their “tokenmaxxing” and excessive AI usage, telling them to focus on what matters: building useful software and achieving company goals. The guidance emphasizes outcomes over maximizing AI token consumption.

Neocloud thinks AI agents - not meatbags - will be its main customers

Quake AI, a neocloud provider, expects autonomous AI agents rather than humans to become its main customers. It is developing APIs to let those agents access its services more efficiently.

🐧 Linux & Open Source

Linux Kernel Developers Considering Removal Of KGDB Debugging Functionality

Linux kernel developers are considering removing KGDB, the GDB-based kernel debugging feature introduced in Linux 2.6.26, after Intel developer Peter Zijlstra posted a patch to delete it entirely. Zijlstra argued KGDB is rarely used, causes serious problems such as NMI-context issues and watchdog timeouts, and is unnecessary given cleaner VM-based debugging via QEMU’s gdb-server. A suggestion was raised to first disable KGDB through Kconfig and remove it in later kernel releases.

System76 Introducing AMD-Powered Linux Small PC

System76 is teasing the Thelio Prime, an AMD-powered Linux mini PC in an 11-liter chassis that launches November 18. It supports AMD Ryzen 9000 CPUs, up to 96GB DDR5 memory and 8TB NVMe storage, with AMD integrated or AMD/NVIDIA discrete GPU options. The chassis is manufactured and systems are assembled at System76’s Colorado facility.

Canonical Reviving Their Immutable Ubuntu Core Desktop

Canonical is reviving Ubuntu Core Desktop, its immutable Linux desktop, after development stalled following its 2023 preview, with a new team and plans to accelerate work for the Ubuntu 27.04 cycle. The effort will initially target browser-centric machines, developer workstations, and gaming machines while addressing challenges including Snap application availability, desktop modularization, permission prompting, and KDE Plasma support. Daily images of Ubuntu Core Desktop are expected during the 27.04 cycle.

CachyOS vs. Fedora 45 vs. Ubuntu 26.10 Performance On AMD Ryzen 9

Phoronix published its first benchmarks comparing CachyOS, Fedora Workstation 45, and Ubuntu 26.10 on an AMD Ryzen 9 9900X3D2 desktop. The test covered dozens of CPU/system workloads, while graphics tests were omitted because Ubuntu 26.10 hit an AMDGPU/Radeon RX 9070 GRE regression that required nomodeset. CachyOS and Ubuntu 26.10 used Linux 7.3, Fedora 45 was on Linux 7.2, and Mesa 26.2 was common across the distributions.

NetBSD Adds Alternative Text Editor To vi & xedit For Easing New Users

NetBSD-current has added the atto text editor to its base system as an alternative to vi and xedit, aimed at easing the transition for new users. Atto is a screen-based editor similar to nano and pico, only 48KB in size, and NetBSD is shipping a modified version with expanded option handling.

X.Org XCB Co-Creator Praises Wayland’s Protocol Design

Jamey Sharp, a longtime X.Org developer and XCB co-creator, praised Wayland’s protocol design after writing his first Wayland client. He said the protocol reflects lessons learned from X and XCB, offering flexibility at the low level while leaving ease of use to higher-level toolkits.

Provenance Time Proposed As A New Timestamp For Linux

Provenance Time (“ptime”) has been proposed as a new Linux kernel timestamp that indicates when a file or directory was created while preserving the original creation date of its contents across file-system boundaries. Request for comments patches enable ptime in the VFS layer and add an initial Btrfs implementation, with proof-of-concept Coreutils patches for stat and cp.

Hyprland 0.57 Released With Compiz Inspired Wobbly Windows

Hyprland 0.57 has been released with Compiz-inspired wobbly windows support, including rules to disable the effect. The Wayland compositor update also adds blur variants, workspace sharing, and other enhancements.

VKD3D-Proton 3.1 Released With New Features Ahead Of Big v3.2 Release

VKD3D-Proton 3.1 has been released by Valve’s Hans-Kristian Arntzen with new features and performance optimizations, serving as the last stable release before the larger VKD3D-Proton 3.2. The update adds descriptor heap support via VK_EXT_descriptor_heap but not enabled by default, KHR_opacity_micromap support for DXR 1.2, Shader Model 6.9, more NVIDIA NVAPI intrinsics, AgilitySDK 619 features, and default view instancing. It also includes performance fixes and game-specific workarounds.

GhostBSD 26.2 Released As FreeBSD 15.1 Powered OS With New Installer

GhostBSD 26.2 has been released, now based on FreeBSD 15.1-RELEASE and continuing to use the MATE desktop atop XLibre by default. The release introduces a new installer with ZFS and network sections plus GELI full-disk encryption support, and it also reduces ISO size, adds a Ctrl+Alt+T terminal shortcut, and enables USB HID support by default.

Linux NTFS3 Hardened Against Corrupt Volumes

Linux 7.4’s NTFS3 driver has been hardened against corrupted or maliciously crafted NTFS volumes, with fixes prompted by fuzzing and KASAN reports. The patches address a possible 64K kernel heap leak, uninitialized page cache contents, a heap overflow, and a mount writing past the MFT record buffer. The update also adds a default-enabled “ugm” mount option; with ugm=0, newly created files omit WSL uid/gid/mode extended attributes.

Retiring BSD-Compress Proposed For Linux PPP Network Code

A patch has been proposed to remove BSD-Compress support from the Linux kernel’s PPP networking code because the feature is unmaintained, orphaned, and largely legacy. The code has a decoder accumulator-width bug on 64-bit systems that can cause a fatal bad-code error and disable compression. The removal patch is pending review on the netdev mailing list.

Wine Merges Support For MESA_map_buffer_client_pointer To Improve WoW64 32-bit OpenGL

Wine has merged support for the MESA_map_buffer_client_pointer extension, improving 32-bit OpenGL support under WoW64 for Windows 32-bit applications and games using 64-bit system libraries. The CodeWeavers-led effort lets Wine provide 32-bit address ranges to the driver for efficient buffer mapping, with the Wine-side merge following the extension’s March merge. The support is expected in Wine 11.21 on Friday; Zink already supports the extension, and a RadeonSI implementation merge request is open.

Emacs celebrates 50th anniversary

EMACS, one of the first full-screen text editors, turns 50; it originated at MIT’s AI Lab in the 1970s from TECO/Control-R and Richard Stallman’s work, with every key bound to customizable macros, and was later rewritten for many operating systems including Mac OS X. The software-sharing community around it helped give rise to the modern free software movement and GNU/Linux, and EMACS remains widely used.

Peel: A free, open source uninstaller and cleaner for macOS

Peel is a free, open-source uninstaller and cleaner for macOS. It is described as a mostly AI-generated tool that is genuinely useful and well-designed rather than low-quality. A linked README explains how Peel was made.

AlpineBSD

AlpineBSD is a work-in-progress operating system based on the NetBSD kernel with an Alpine Linux-inspired userspace, aiming to minimize AI-generated software; its base system and main package repository contain no LLM output. It currently lacks an installer but offers a high-level installation guide, and it ships apk-tools and BusyBox while currently using s6, with OpenRC planned as an alternative.

CaseCite is a source-available, self-hosted legal research platform offering eight tools for document indexing, cited Q&A, contract review, drafting, and research into cases, judges, and authorities. Documents, indexes, chat history, and analyses stay on the user’s infrastructure, while AI features send only relevant content to user-configured providers under their own API keys with no telemetry to CaseCite maintainers. It is licensed under Elastic License 2.0 and free to self-host for internal firm use and modify.

Popkern – open-source CAD kernel with G2 corner blends, browser IDE

Popkern is an open-source, free-forever parametric B-rep CAD kernel that lets users build and edit watertight solid models using short Rhai scripts in the browser or locally, with STEP import/export and migration-style project histories. It currently supports booleans, smooth-corner fillets/chamfers, sweeps, revolves, lofts, shelling, constrained sketches, and a feature tree with undo, though it is still early. The AI-agent-built kernel is heavily test-verified against OpenCASCADE and licensed AGPL-3.0 with an exception making users’ scripts, models, and STEP files their own.

Astroterm: A Planetarium for Your Terminal

Astroterm is a terminal-based star map written in C that displays real-time positions of stars, planets, constellations, and moon phases using ASCII-rendered visuals. It lets users configure sky views by date, time, and location and is available through several package managers or by building from source.

A report from the LPC sched_ext microconference

The Linux Plumbers Conference report on the sched_ext microconference covers six talks and an opening discussion on future directions: moving more scheduling logic into BPF, composability, and generating BPF code from Rust. The main topics were priority inversion—including proxy execution compatibility with sched_ext, lock-holder/waiter preemption, and time-slice extension—paravirtualized host-guest scheduling with sched_ext as the host policy engine, and device/topology-aware placement using Intel HFI, accelerator locality, and cache topology. Proxy execution support is queued for Linux v7.4, allowing distributions to enable CONFIG_SCHED_PROXY_EXEC and CONFIG_SCHED_CLASS_EXT together.

Redox OS ring buffers project summary

Redox OS has published a report on its grant-funded effort to implement io_uring-like I/O using ring buffers, with the current phase finishing in September. The team implemented ring-buffer machinery for supported system services and drivers, completed and demonstrated an io_uring/liburing-like API for applications, and developed support crates including an async executor for drivers. It also reported performance improvements enabling faster switching between applications and their system services.

💻 Software & Development

The Future Trait Has A Missing State

The article examines async Rust’s Future::poll semantics, noting that futures are intended to resolve once to Poll::Ready and that polling again afterward is unsupported and often panics. It discusses the MaybeDone wrapper, which stores either a future or its completed output, and argues that being able to poll a future again after Ready can be useful.

Who’s 0xabadbabe and why? (2011)

While simulating full x86 systems, an engineer traced a Windows 7 blue screen to a bug in a simulated USB-disk device. The mysterious write of 0xABADBABE to a reserved EHCI register turned out to be an intentional Windows debugger breakpoint, not the failure cause. The real bug was that the simulated EHCI controller returned USB data-structure pointers outside the driver’s mapped memory.

Scalable Vector Grief

The article reports that the author tested eight popular SVG sanitizers and bypassed six, including mattkrick/sanitize-svg, Borewit/svg-sanitizer, bgalek/safe-svg, and alnorris/SVG-Sanitizer. It demonstrates JavaScript execution in Firefox via a response header even with harmless XML, and in Safari by importing a GIF as SVG. It also details bypass techniques involving XHTML namespace switching, foreignObject casing issues, and animate/animateTransform payloads.

StackOverflow now gets fewer than 1000 questions a month

StackOverflow now gets fewer than 1,000 questions a month. The provided article content was inaccessible because it only displayed a JavaScript and cookies requirement.

GitHub scales infrastructure to prevent outages

GitHub is re-architecting its platform and has added 4 million CPU cores, 160 PB of high-speed storage, and 50 Tb/s of network capacity to handle rapid growth and prevent outages. The company reported 7.4 billion commits in September, up from 2.9 billion in August, driven largely by agentic and AI-related demand.

A proof-of-concept formalization of the IRS federal tax forms

A proof-of-concept project has formalized IRS federal tax forms using “vibe-coded” software, guided by the principle that such tools should be verifiable and let users trace interpretations back to authoritative source documents. The project is a passive, back-burner effort explicitly not for real use, but its creator argues something like it should exist for the public good.

noalloc: Tooling and libraries to enforce zero heap allocations in Golang

noalloc is an experimental, AI-written tool and library set for Go that helps enforce zero heap allocations by letting developers mark functions with //noalloc:root; its CLI then checks the full call tree and reports every line that allocates. It also provides zero-allocating, fixed-sized data structures and algorithms, including a JSON encoder and decoder, to simplify writing non-allocating code. The project is used in production at The Consensus and requires Go 1.27 or later.

Scaling Hierarchical Networks

Researchers introduce hierarchical networks (H-Nets), a general architecture aimed at learning end-to-end from raw bytes without hand-crafted tokenizers, in line with the “bitter lesson.” Scaling results show H-Nets can deliver up to 3–5x compute efficiency through learned tokenization, while a more flexible 2-stage H-Net outperforms a matched open-source model with 10x less compute and data. The work also reports novel behaviors such as in-context tokenization and early promise across modalities including biology and vision.

SlideBySide – Present slides without sharing your screen

SlideBySide enables presenters to show slides without sharing their screen while keeping everyone on the same page. It also lets cohosts present together and jointly control the current slide, with additional presentation collaboration features. The article notes that building a DIY version is a hassle because maintaining an online version is difficult.

CNC Stone Carving

The article explains how CNC stone-carving technology can help revive Gothic architecture by rapidly producing ornate details that once took skilled carvers months. It describes computer-controlled, multi-axis machines that use diamond tools to carve stone and can work unattended, cutting time and cost. The Carmelite Monks use Prussiani CNC machines, channeling young people’s computer skills toward church construction.

Yavashark: I’ve Built a JavaScript Engine Every Day for 950 Days

The article announces Yavashark, a JavaScript engine that its creator has built every day for 950 days. The author invites feedback and discussion, saying they enjoyed the project and hope readers did too.

An architectural comparison of Obsidian Canvas and Heptabase

Obsidian Canvas uses a file-centric architecture, storing visual arrangements in open JSON .canvas files so connections are portable but unindexed in the vault graph. Heptabase uses a card-centric architecture with a central card library, where whiteboards are non-owning views and connections are indexed across cards but locked into its proprietary ecosystem. The comparison argues these data models shape spatial thinking, graph indexing, and data portability.

Nasscad – local-first 3D CAD for browser and Windows

Nasscad is a local-first 3D CAD tool for the browser and Windows, with a welcome tutorial, project import/open, and NassScript demos. It supports STEP AP203/AP214/AP242 import and export, assembly browsing with retained part names, placements and colors, and PMI/GD&T display when supported by the source. Projects can be saved as NASSCAD JSON and exported to other tools, with MEDUSA native processing on Windows and large-assembly support via STEP streaming and Turbo.

Squishy 3D helpers powered by Three.js

Plushies is a Three.js-powered tool for creating soft 3D plush characters on the web, with 17 silhouettes that inflate into pillows featuring rolled seams and fur. The characters can be used as AI-helper cursors, customized with live code updates, and installed via npm, while the page also covers real recycled-felt plushies, care and repair guidance, and worldwide shipping over €40.

Billion-query-per-second lookup tables in C++

The article describes using compile-time perfect hashing in C++ to map fixed strings to values more efficiently than std::unordered_map, generating a collision-free hash function during compilation via a greedy algorithm. It also explains that runtime lookup uses a cheap hash computation and vectorized 16-byte string comparisons to reduce branching and instruction count.

Amazon Software Engineer AI Guidelines

Amazon engineering leadership has updated its Software Development Engineer role guideline, reaffirming core SDE principles while acknowledging AI’s growing impact on software development. The company introduced the idea of the “Renaissance Developer,” who combines AI tools with human judgment and domain expertise. Amazon said it believes AI’s effect on software engineering is still early and that engineers must embrace the change and keep learning.

A Lego-like composable Framework for RSI

GEAR is an open general evolutionary architecture for recursive self-improvement of LLM agent harnesses. It uses a two-level meta-learning process: an inner loop refines reasoning, planning, and actions within tasks, while an outer Meta Agent identifies cross-task patterns and proposes reusable harness improvements; evolutionary search is modularized into replaceable components, with Staged GEPA as one instantiation. Reported same-model comparisons show improved task pass rates and objective completion on AutomationBench marketing/support and Terminal-Bench 2.1.

Solving Postgres’ Count Problem

ParadeDB optimized PostgreSQL COUNT queries by partitioning segments by ctid, allowing visibility checks to be performed up front and greatly reducing page accesses. The article also tests filtered and sorted BM25 search queries, reporting ParadeDB is up to 66 times faster and explaining what these benchmark differences mean for search engine selection.

Quoting Ted Sanders

Ted Sanders explains that while LLM performance is multidimensional, model size explains much of the variation: larger models are generally more capable and reliable, cost more per token, and better recall esoteric facts. OpenAI’s models are ordered Luna, Terra, Sol, and Astra, and version numbers indicate release order, relationships, and magnitude of improvement rather than a fixed performance ranking.

GitHub Octoverse 2026

At GitHub Universe, EVP Jay Parikh announced the 2026 Octoverse report, highlighting agentic engineering’s impact on GitHub. Merged pull requests tripled in less than a year to 220 million per month, commits more than tripled to 7.38 billion per month, and new repository creation doubled to 30.3 million per month. The report also adds industry breakdowns and counts repositories involving interoperability or reverse engineering, with methodology notes on country and sector classification.

Anthropic engineer declares the coding interview dead

An Anthropic engineer has declared the coding interview dead, asking why candidates should be given tests that AI can already ace. The remark raises questions about how technical hiring should adapt as AI coding capabilities improve.

How Google fumbled the frontend with Angular 2

A talk at All Things Open examined how Google mishandled the frontend upgrade to Angular 2. It presented the episode as a case study in how not to upgrade a software package.

🛍️ Consumer Tech & Business

How to Watch Liverpool vs Chelsea for Free (October 28th, 2026)

Liverpool host Chelsea in the Carabao Cup fourth round at Anfield on October 28, 2026, with kickoff at 8:00 PM local time. UK viewers can watch free via ITV1/ITVX, Scottish viewers via STV/STV Player, and viewers abroad can use a UK- or Scotland-based VPN connection to access those broadcasts.

NordVPN vs Proton VPN Black Friday deals 2026 are live: up to 78% OFF

NordVPN and Proton VPN have launched their 2026 Black Friday deals, offering discounts on long-term premium subscriptions. NordVPN advertises up to 78% off selected two-year plans, while Proton VPN offers up to 75% off depending on the plan. Both providers include a 30-day money-back guarantee.

iJustine accused of “engagement bait” after tone deaf comment about iPhone Duo

iJustine faced backlash and was accused by Mark Gurman of “engagement bait” after implying iPhone Duo users may be deliberately breaking their devices. Her comment came as owners reported cracked internal screens, white lines, and dead pixels on the new $2,000 foldable. Apple acknowledged the issue and offered replacements, but Gurman said stores lacked enough stock.

Firefox brings back Bad Luck Brian and other meme legends to resist big tech

Firefox has launched a nostalgia campaign featuring early internet meme stars such as Bad Luck Brian, Overly Attached Girlfriend, and the “Apparently” kid to criticize big tech surveillance, intrusive ads, and built-in AI users cannot opt out of. The browser is positioning itself as a return to a more private, personal internet, and the campaign video drew over 923,000 YouTube views in its first two days while sparking Reddit nostalgia.

Discord makes public communities searchable, users immediately look for a way to opt out

Discord will create searchable web pages for public Discoverable Servers with at least 1,000 members starting November 16, potentially exposing summaries of server conversations in web search. Private servers, DMs, and group DMs are unaffected, but users in most regions must opt out to prevent eligible messages from being indexed, while some regions have the feature off by default. The change has prompted immediate user backlash and advice to adjust privacy settings before rollout.

X adds an intrusive promotional animation of an arm reaching across the screen to direct users to…

X is using an intrusive animated arm that reaches across the screen to highlight the “promote” button for verified or Premium users, drawing negative reactions from users who called it jarring and Halloween-like. The paid promote feature extends a post’s reach to larger audiences based on location, interests, or demographics, including non-followers.

Amazon Kindle Paperwhite and Kindle Colorsoft review: premium feel, lightweight, nice colors, and…

The new 7-inch Kindle Paperwhite and Kindle Colorsoft feature refreshed designs, optional page-turn buttons via an $80 cover limited to Signature Editions, and new device colors on Signature models, with prices matching the 2024 models after Amazon’s earlier price increase. They are solid new picks—especially for the page-turn buttons—but upgrading mainly makes sense for readers who want that feature, while the Paperwhite remains a good first Kindle unless a color screen is essential.

Samsung unveils the $229 Galaxy Buds On, its first open-fit, C-shaped cuff earbuds to compete wit…

Samsung unveiled the $229 Galaxy Buds On, its first cuff-style earbuds, featuring a C-shaped clip-on design that hugs the earlobe. The open-ear design leaves the ear canal unobstructed for ambient awareness, and the black earbuds go on sale Nov. 6.

Etsy reports Q3 revenue up 11% YoY to $692.5M, above $668M est., net income of $853.5M after sell…

Etsy reported Q3 revenue up 11% year-over-year to $692.5 million, beating the $668 million estimate, with net income of $853.5 million. The company saw its first year-over-year increase in active buyers in two years and its fastest active-seller growth in two years. Etsy cited early signs of success under CEO Kruti Goyal as personalization and younger-shopper targeting efforts begin to pay off.

Amazon unveils the $500 Echo Show 16, the $200 Echo Time alarm, and the $200+ Alexa Photo Frame, …

Amazon unveiled a new Alexa-focused device lineup, including a $500 16-inch Echo Show, a $200 Echo Time bedside alarm, a $200+ Alexa Photo Frame, and refreshed Echo speakers. The products are intended to extend Alexa’s presence throughout the home as Amazon faces renewed competition from Apple.

How the UK’s Slough, a town on the borders of west London, became the largest data center hub in …

The provided content is not a news article but an FT.com security verification page showing Error Code CG000/403 and Request ID a51aa25b3dcdae8e. It includes a link to help.ft.com and an apology for inconvenience, with no substantive information about Slough’s data center hub.

Get Ready for Touch on Mac

Apple is preparing developers for touch input on Mac through online Touch on Mac Group Labs with Apple engineers and designers. Developers can download the Xcode 27.3 beta SDK to build, run, and test apps, with documentation and design resources coming soon.

The first wired XR glasses powered by Android XR

Google and XREAL launched XREAL AURA, the first wired XR glasses powered by Android XR. The lightweight glasses feature transparent lenses, a 70-degree field of view, and a wired compute puck with the Snapdragon Reality Elite platform, enabling spatial apps, Gemini assistance, and laptop mirroring for work, watching, and play.

Formlabs Studio 1 $699 Resin Printer

Formlabs is offering the Form Studio 1 Founders Edition, a $699 professional-grade resin 3D printer (excluding sales tax) aimed at making resin printing more affordable. It combines a long-lasting print engine and open platform with razor-sharp detail, mess-free resin handling, and a UL GREENGUARD Gold Certified workflow using Alpha Resins and a VOC Extractor, with most parts printable in under two hours.

Alien 3 Is Getting a Second Chance on the Amiga CD32, More Than 30 Years Later

A developer known as Mister_Grave has released Alien 3 CD32 Edition, a substantial fan overhaul of the 1993 Commodore Amiga game. The project rebuilds the game engine and uses the Amiga CD32’s AGA hardware to add 64-colour visuals, parallax scrolling, and elements from the Sega Mega Drive version. It aims to deliver the closer-to-16-bit Alien 3 experience the developer believes Amiga players should have received originally.

Alphabet Announces Third Quarter 2026 Results [pdf]

Alphabet announced its third quarter 2026 results. The announcement was published as a PDF.

Nintendo’s worst nightmare begins as Switch 2 games start running on PC and Mac

Days after the Switch 2 jailbreak became public, developers have demonstrated Switch 2 games running outside Nintendo’s console. Mario Kart World was shown on an iPhone 18 Pro using a private nxnative fork, while Splatoon Raiders reportedly booted on PC via a modified Eden emulator with poor performance and missing features. The demonstrations use private builds, and the results have not been independently verified through a public release.

Mac Studio M5 Ultra 512GB Officially Released

The article titled “Mac Studio M5 Ultra 512GB Officially Released” is largely an Apple FAQ covering Mac setup, migration, education discounts, delivery options, external displays, and payment methods. It does not provide substantive details about the announced Mac Studio M5 Ultra 512GB.

MacBook Pro PROLED XDR Display Touch

Apple unveiled new 14- and 16-inch MacBook Pro models with a PROLED XDR tandem OLED display and Touch on Mac for the first time, offering up to 2000 nits brightness, a 120Hz variable refresh rate, 10,000,000:1 contrast, M5 Pro/M5 Max chips, up to 128GB unified memory, up to 16TB storage, and up to 24 hours of battery life in a thinner, lighter design. Pre-orders start October 28 with availability beginning November 6, though one source excerpt did not include pricing or availability details.

Memory Chip Shortage to Persist Through 2030

SK Hynix says its DRAM and high-bandwidth memory capacity is effectively sold out through 2029, with shortages expected to persist until 2030 as demand outpaces supply. Customers have locked in more than 50% of 2027–2028 capacity through long-term agreements, and new capacity will take years to come online. The shortage is expected to keep memory prices elevated and hurt consumers and PC/smartphone markets, though AI infrastructure demand should offset those declines and benefit SK Hynix’s bottom line.

Old.reddit.com Gone

A post reports that old.reddit.com appears to be gone, though the author notes it may only be their experience. The author says they feel free as a result.

Tesla Semi catches fire on I-880

A Tesla Semi caught fire in the southbound lanes of Interstate 880 in Hayward, California, on Tuesday evening without a crash, and the driver escaped unhurt. The truck’s lithium-ion battery burned long enough to keep the southbound freeway closed for about 10 hours, and the cause is under investigation. It was the second Tesla Semi fire to close a California interstate, and Tesla has not commented.

Dropbox lays off 9% of workforce

Dropbox announced Wednesday that it is laying off roughly 9% of its workforce. The company expects to spend between $27 million and $33 million on the layoffs, largely for severance payments and employee benefits.

Valve Responds to Developers Fearing AI Slop and Clones

Valve advised developers worried about AI clone games to focus on community building and consistent updates, saying player trust and a distinct, continuously evolving game identity are harder to copy than a single mechanic. At the October 27 Steam Indie Connect event, Valve also discussed Steam’s growth, launch preparation, and new tools. Responding to AI-cloning extortion threats, Valve noted that clone games predate AI and that AI has accelerated them.

Suffering as a Feature

A website visualizes extreme U.S. wealth inequality as an intentionally arduous, linear scroll from the poorest to the richest Americans, forcing users to endure inconvenience to grasp the scale. The author argues this “suffering as a feature” design conveys the reality of poverty, noting that all U.S. poverty could be eliminated for $177 billion but society chooses not to. The article also discusses support for a 2026 wealth inequality video series and a printable foldable version of the graph.

‘Forced out’: HOA fees to rise to $2,500 a month

An article titled “‘Forced out’: HOA fees to rise to $2,500 a month” indicates that homeowners’ association fees are set to increase to $2,500 monthly. However, the provided page contained only an Error 451 notice saying the website is unavailable outside the United States, so no further details were accessible.

Man spends 35 years collecting corporate waffle, turns it into an app

A man spent 35 years collecting corporate waffle—buzzword-heavy business jargon—and has turned the archive into an app. The app is pitched as a way to generate useful-sounding corporate language, or “leverage synergies,” without having to think of something to say.

Why roaming packages have become an anachronism in the age of app-first travel

Traditional roaming is increasingly outdated for app-dependent travel, where constant connectivity is essential but operators impose rigid zones, limited data allowances, and opaque pricing. Travelers are turning to alternatives such as travel eSIMs, which let them choose trip-specific plans and see costs in advance. Services like Yesim offer a more flexible, transparent alternative to home-operator roaming packages.

AI slows growth for Infosys and keeps it afloat at the same time

Infosys predicts slower growth as savings from AI-driven automation pressure the pricing of extended deals. At the same time, AI is helping keep the company afloat, according to the article.

🌍 Science, Health & Society

Russia desperate to camouflage its data centers to protect them from Ukraine’s drones

Russia’s Digital Development Ministry has approved guidelines for data center operators to reduce aerial visibility, install anti-drone nets and electronic warfare, and potentially move new facilities underground amid increasing Ukrainian drone attacks. New data centers would also need to protect server halls, cooling systems, and power supplies from shockwaves and shrapnel. The measures could raise service costs by about 10% on top of inflation, while experts warn a major outage could destroy 87% of a facility’s value.

Low cholesterol targets cut cardiovascular events by 33%

In the Ez-PAVE randomized trial, patients with cardiovascular disease assigned to an LDL target of <55 mg/dL had a 33% lower rate of cardiovascular events over three years than those targeted to <70 mg/dL (6.6% vs 9.7%). Only 61% of the intensive group reached <55 mg/dL, suggesting the benefit ceiling may be higher with better access to newer drugs. The trial did not establish a reduction in cardiovascular deaths, though fewer heart attacks and procedures accounted for much of the benefit.

Outrage after airport self-service checkout asks for 18% tip

A self-service checkout at Newark Liberty International Airport prompted outrage after asking customers for an 18% tip despite no worker being present. The Port Authority of New York and New Jersey said the concessionaire OTG will remove the tipping prompt immediately and ensure other kiosks do not ask for tips. The incident comes amid broader consumer frustration over tipping culture, with most respondents in a recent study calling current practices ridiculous.

A Health Update

The author recounts his chemotherapy experience, starting with weekly spinal methotrexate injections via lumbar puncture that caused severe spinal headaches, followed by Bendamustine and Rituximab infusions on a 28-day cycle. Side effects included a metallic taste, changed eating habits, lifeless hair, fatigue and muscle loss, and about 50 pounds of weight loss, though he avoided nausea and hair loss and typically felt better by the middle of each cycle.

Mecca Defence Alliance

Pakistan, Saudi Arabia, and Turkey signed the Mecca Defence Alliance into force on 7 August 2026, establishing an intergovernmental military alliance headquartered in Riyadh with a collective-security commitment modeled on NATO Article 5. The pact, formerly the Mecca Joint Defence Agreement, builds on the 2025 Pakistan-Saudi Strategic Mutual Defence Agreement and was announced amid the 2026 Iran war. Its agreement has been submitted to Turkey’s parliament for ratification, though the full text has not been released.

Rising rates of colorectal cancer found in nearly all age groups

A study in NEJM Evidence found that colorectal cancer incidence is rising across nearly all age groups up to 65–69, with the overall age-adjusted rate increasing in 2023 for the first time in more than 20 years. The sharpest rise was among adults aged 45–49, likely reflecting expanded screening, but increases in older age groups suggest unknown risk factors may be offsetting screening gains. Colorectal cancer remains the second leading cause of U.S. cancer deaths overall and ranks first among adults younger than 50.

Spain high-speed train crash kills 68

Spain has begun three days of mourning after two high-speed trains collided south of Madrid near Toledo, with reports citing at least 68 dead while an earlier account said at least 10 and warned the toll could rise. A Madrid-to-Seville train struck the rear of another stopped on the same track, derailing carriages and involving RENFE and IRYO services, as rescue workers searched the wreckage and leaked audio suggested a train was cleared to move despite another stalled ahead. A former minister was among the dead, King Felipe VI visited a hospital, rail services between Madrid, Ciudad Real and Andalusia were suspended, and it is Spain’s second major rail tragedy of 2026 after a January crash killed 45.

Big Tech’s thirst for water is sparking a global resistance

The rapid expansion of AI has sharply increased Big Tech’s water use, with one estimate putting initial ChatGPT training at 185,000 gallons and hyperscale data centers consuming up to 1–5 million gallons a day. Google and OpenAI have disputed broader water-use estimates as too narrow, while independent researchers warn that large AI data centers compete with communities for potable water. The issue has fueled public resistance, including in Chile, where Google sought approval for a $200 million data center amid water-focused protests.

Spiders Spin ‘Silk Dope’ into Nature’s Toughest Material

Researchers have mapped how spider silk proteins in liquid “silk dope” first form semiliquid condensates before solidifying into fiber, offering an atomic-level view of spiders’ silk-spinning process. The findings could help explain how nature produces silk tougher than Kevlar and stronger than steel, and inform efforts to recreate it for materials such as seat belts, helmets, and brain-cell scaffolds. Lab-made spider silk has lagged because the spinning mechanism was poorly understood.

It’s now official, tech is the new tobacco

Meta is facing another lawsuit alleging its addictive products harm minors, raising difficult questions about causality between social media use and youth mental health. The article argues tech companies are following tobacco’s playbook by shaping scientific research, and newly released internal Meta documents reportedly show plans to discredit critics, limit internal youth-mental-health research, and influence National Academies work. It also cites research finding that at least half of high-profile social media studies had disclosable ties to tech companies, many undisclosed.

Meta censored the No Kings Instagram account during a nationwide campaign

Meta suspended the No Kings Coalition’s Instagram account on 17 October during a nationwide campaign encouraging early voting for the 3 November midterms, when nearly 3,000 events were planned. The account became publicly visible again after about 90 minutes, but organisers could not post until Meta restored access at 5:33pm. Meta initially blamed excessive posting and commenting, then said software using Claude AI was reading Instagram at four times the normal rate; it admitted the suspension was wrong, but its changing explanations did not account for what happened.

Germanys’ AfD in Saxony-Anhalt: Porno, Putin, Criminals, Nazi Songs, Corruption

Thomas Klikauer’s article reports that the AfD won 44% of the vote and 39 of 83 seats in Saxony-Anhalt’s September 2026 state elections, with an overwhelmingly male parliamentary group. With three seats short of a majority, the AfD relied on BSW support to elect Tobias Rausch as speaker and plans for BSW votes for Ulrich Siegmund as state premier. Klikauer characterizes the AfD deputies as extremists, opportunists, and criminals, citing nepotism, corruption, neo-Nazi ties, antisemitism, and pro-Putin sympathies.

Amazon rainforest cut down to build highway for COP climate summit (2025)

A new four-lane highway is being built through protected Amazon rainforest in Belém, Brazil, to ease traffic for the COP30 climate summit expected to draw more than 50,000 people including world leaders. The state government calls the road “sustainable,” but locals and conservationists say the deforestation, lost livelihoods, and fragmented wildlife habitat contradict the purpose of a climate summit.

Trump says he’s ending restrictions on climate-warming methane from oil drilling

President Trump said he is terminating a Biden-era rule restricting methane emissions from oil and gas drilling, claiming it will save $45 billion and unlock 210 million barrels of oil. The rollback would eliminate the EPA’s “Super Emitter Program” and set less restrictive standards for small wells. Environmental groups criticized the move, saying it gives companies a “free pass” to pollute.

Kentucky man becomes the first humanist granted religious exemption from AI

A Kentucky man has become the first humanist to be granted a religious exemption from AI. Further details were unavailable because the article content required JavaScript and cookies to access.

Please Let the Internet Die

The article argues that the current internet is in advanced cultural decay, citing lawsuits against Meta and Google over privacy and mental-health harms, school smartphone bans, under-16 social media restrictions, and generative AI flooding feeds with low-quality content. It says bots and monetized rage bait increasingly dominate online interaction, making the web feel manipulated and exploitable. It introduces John West’s book The Internet Will Die, and So Will You as a treatise confirming these suspicions.

The Brutal Harvard Class That Suddenly Everyone Wants to Take

Humanities 10, a demanding Harvard course, has become unexpectedly popular amid concerns about the decline of the humanities. In a recent session, professor Samantha Matherne discussed the philosophical underpinnings of Jane Austen’s “Emma,” connecting it to Adam Smith and Immanuel Kant.

I’m going on strike this November

The article calls for a strike in November and instructs participants to notify their employers that they are going on strike, optionally linking to the thread and explaining their situation. It says people should not debate whether they can participate, arguing that the action will work if enough people join.

Study: Schoolwide phone bans show no significant impact on mental health

A study found that schoolwide phone bans show no significant impact on mental health. Further details were not available in the article content provided.

Barcelona unrest: how to fabricate a protest in the age of TikTok

TikTok accounts linked to the anonymous Blocus platform promoted an October 24 Barcelona gathering that turned into looting and clashes with police, resulting in 130 arrests and 14 injured officers. The accounts were later deleted as new Blocus-labeled profiles called further gatherings across Spain and Europe. Spanish politicians traded blame over the violence amid speculation about far-right or foreign involvement, though officials said there was no indication of Russian interference.

Spain train crash: ‘At least 10 dead’ in high-speed collision

A high-speed train collision in Spain has reportedly killed at least 10 people, according to the article title. The article body could not be accessed because it displayed device verification errors, so further details were unavailable.

Wealth Inequality in America drawn to scale

The article uses a to-scale scrolling graph to illustrate extreme U.S. wealth inequality, showing that the top 1% is so wealthy it takes minutes of scrolling to reach. It notes that the median household’s wealth equals about a 9-inch stack of $100 bills, while the average top-1% household’s equals about 145 feet, with the top 0.01% extending even further.