Monday, 06 April 2026
Microsoft calls Copilot 'for entertainment only', Germany IDs the REvil ransomware kingpin, and Gemma 4 lands on iPhone
Today's Lead
TechCrunch
Copilot Is 'For Entertainment Purposes Only', per Microsoft's Terms of Service
Microsoft's terms of service classify Copilot as a tool "for entertainment purposes only" — a stark contrast with the company's aggressive marketing of the product to enterprises and professionals. The disclosure highlights a pattern across the AI industry in which vendors embed protective disclaimers acknowledging the unreliability of their models while simultaneously promoting them for high-stakes use. The gap between contractual liability language and commercial messaging is becoming increasingly difficult to ignore as AI assistants take on real business workflows.
Also today
Krebs on Security
Germany Doxes 'UNKN,' Head of Ransomware Gangs REvil and GandCrab
Germany's Federal Criminal Police publicly identified Daniil Maksimovich Shchukin, a 31-year-old Russian national, as "UNKN" — the operator behind the GandCrab and REvil ransomware gangs responsible for at least 130 cyberattacks and over €35 million in economic damage. Shchukin is credited with pioneering the double-extortion model, in which victims pay separately for a decryption key and for stolen data not being published. The identification is notable for connecting a long-pseudonymous threat actor to a real identity; Shchukin is believed to be in Krasnodar, Russia and is unlikely to face immediate prosecution.
Read →Simon Willison
Google AI Edge Gallery Brings Gemma 4 Locally to iPhone
Google released an official iOS app — Google AI Edge Gallery — for running Gemma 4 models entirely on-device, with the 2.54 GB E2B variant performing both quickly and usefully in practice. The app supports image question-answering, 30-second audio transcription, and an interactive tool-calling demo, making it the first official mobile app from a major AI lab aimed at on-device model experimentation on iPhone. Significant limitations remain: conversation history is ephemeral and the app froze during follow-up prompts in early testing.
Read →lalitm.com
Eight Years of Wanting, Three Months of Building with AI
Lalit Maganti chronicles how AI coding agents helped him complete syntaqlite — a high-fidelity SQLite linting and parsing toolset he had wanted to build for eight years — in three months. The piece offers unusually candid analysis of where AI assistance fails: architectural design, where the low cost of AI-assisted refactoring encourages deferring key decisions until the codebase becomes incoherent, and where problems with no objectively checkable answer expose the limits of code-generation tools. Maganti's conclusion — that AI is "autocomplete on steroids" rather than a design partner — is backed by a detailed account of discarding an entire AI-generated prototype and rebuilding from scratch with much more human-in-the-loop oversight.
Read →TechCrunch
In Japan, the Robot Isn't Coming for Your Job — It's Filling the One Nobody Wants
Japan is deploying physical AI robots at commercial scale to fill vacancies that humans actively avoid — a direct consequence of acute demographic labour shortages rather than a drive to cut costs. The framing inverts the dominant Western anxiety about automation: robots here are a supplement to a shrinking workforce rather than a replacement for an existing one. Major investors including Salesforce Ventures and Woven Capital are backing the trend, and Japan's model is being watched as a template for other ageing economies.
Read →sschueller.github.io
Why Switzerland Has 25 Gbit Internet and America Doesn't
Stefan Schüller argues that Switzerland's superior fiber internet — reaching 25 Gbit speeds — is the product of intelligent regulation rather than free-market competition. The Swiss model treats physical fiber as a shared neutral asset with mandated open access, allowing multiple ISPs to compete on services over the same infrastructure instead of building duplicate networks. The article contrasts this with the American territorial-monopoly approach and the German model of redundant parallel buildout, making the case that separating natural-monopoly infrastructure from competitive services is the policy variable that actually determines consumer outcomes.
Read →The Document Foundation
LibreOffice — Let's Put an End to the Speculation
The Document Foundation published a frank acknowledgement of past governance failures — including selective brand licensing, conflicts of interest from company representatives on the board, and the 2019 creation of an unofficial parallel organisation — alongside a set of remedial measures. New policies include a Code of Ethics, updated Conflict of Interest Policy, Community Bylaws, and stricter procurement procedures. The statement is a direct response to public criticism from Collabora and other contributors, and positions the foundation ahead of anticipated growth in government adoption of LibreOffice.
Read →jsnover.com
Microsoft Hasn't Had a Coherent GUI Strategy Since Petzold
Jeffrey Snover, creator of PowerShell, argues that Microsoft has lacked a coherent Windows GUI strategy since Charles Petzold's Win32 era — producing instead a graveyard of competing frameworks (MFC, COM, WPF, WinRT, UWP, WinUI, and more) driven by internal politics rather than platform vision. The recurring pattern is a cycle of conference announcements, developer investment, and then quiet abandonment — Silverlight being the canonical example. Snover attributes the fragmentation to a structural division between the Windows and .NET organisations and sees it as the primary driver pushing developers toward cross-platform solutions like Electron.
Read →pscanf.com
Can We Measure Software Slop? An Experiment
Inspired by growing concern about AI-generated code flooding repositories, this post proposes a working definition of "software slop" as code lacking human attention rather than code lacking quality per se, and introduces a "Slop-O-Meter" that combines an attention cost estimate (lines weighted by file type) with attention spent (GitHub review and commit signals). The framework produces reasonable results in standard cases but generates false positives for projects with non-traditional workflows, such as SQLite's Fossil-based development. The experiment surfaces an important distinction: rushed human code and vibe-coded AI output may look identical to automated metrics.
Read →Michael Stapelberg's Blog
Stamp It! All Programs Must Report Their Version
Michael Stapelberg makes the case that embedding and surfacing VCS revision information should be treated as a non-negotiable engineering practice: stamp the revision at compile time, ensure it survives the build, and expose it through --version flags, logs, headers, and UIs. Using the i3 window manager and Go's built-in stamping support as examples, he shows the operational value during debugging and incident response is disproportionate to the implementation cost. Practical tooling, including a Nix overlay for Go projects, accompanies the post.
Read →