Today’s top stories include Linux 6.15 adding Apple Silicon support, AWS transforming VMware migration from weeks to minutes with AI, and a groundbreaking personalized CRISPR therapy curing a baby with a rare genetic disorder. Additionally, Microsoft faces EU delays, UK approves a Green Belt data center, and AI cybersecurity threats continue to rise.
▶️ Internet Infrastructure
Linux 6.15 Adds Apple Silicon Support and Upstream Graphics APIs
Linux 6.15 introduces Apple Silicon support with new drivers and upstreamed graphics API, enabling OpenGL, OpenCL, Vulkan; Fedora Asahi Remix 42 released; further upstreaming and distro support guidelines ongoing.
- Linux 6.15 release includes Apple Silicon support with new drivers for ADP display controller and Z2 touchscreen digitizer, enabling Touchbar support on M1/M2 MacBook Pros
- Upstreamed graphics driver userspace API (uAPI) into the kernel, allowing OpenGL, OpenCL, Vulkan support for Apple Silicon in Mesa; forks of Mesa, virglrenderer, and Flatpak runtime will be sunset in Fedora Asahi Remix 43
- Additional upstream patches include PCIe support for T6020 (M2 Pro), support for TAS2764/TAS2770 speaker amps, and ongoing efforts for SMC and SPMI controllers, supporting system shutdown, GPIO, hardware sensors, and USB Power Delivery controllers
AWS Transform Accelerates VMware Migration from Weeks to Minutes
AWS claims its AI-driven “AWS Transform” tool accelerates VMware workload migration from weeks to minutes, addressing legacy issues and supporting rapid cloud re-platforming amid rising subscription costs.
- AWS criticizes VMware as a legacy platform, claiming it has become “crusty like a mainframe” in 2025
- AWS introduced “AWS Transform,” an AI-powered tool that automates migration of VMware workloads to AWS cloud, reducing migration time from weeks to minutes
- The tool generates migration wave plans for 500 VMs in 15 minutes and performs networking translations up to 80x faster than traditional methods; supports converting VMware networking and firewall rules into AWS equivalents
Microsoft Misses EU Azure Deadline; Seeks Alternative Solutions Amid Regulatory Pressure
Microsoft missed the May 2025 deadline for a specialized Azure version for EU hosters, with plans to propose alternative solutions within two months amid regulatory and legal pressures.
- Microsoft failed to deliver a specialized Azure version for EU cloud providers by the May 2025 deadline, risking legal action if a “commercially equivalent solution” is not proposed within two months.
- The project, Azure Local (formerly Azure Stack HCI), aimed to support multitenancy, unlimited virtualization, multi-session VDI, pay-as-you-go SQL Server licensing, and extended security updates, but was delayed due to underestimated engineering efforts.
- Microsoft and CISPE agreed to abandon the original Hoster Product and pursue “Plan B,” focusing on alternative solutions to meet fair licensing commitments; the ECCO report criticizes Microsoft’s shortcomings and licensing practices harming European CSPs.
Pure Storage Unveils AI-Ready Platform for Life Sciences Data Innovation
Pure Storage’s unified platform enables life sciences organizations to bridge research, development, and healthcare data workflows, supporting AI, unstructured data, and containerized applications with resilient, scalable storage.
- Pure Storage offers a unified data platform designed to support AI workflows, IP protection, and accelerated discovery in life sciences
- The platform handles demanding workloads such as AI pipelines, unstructured datasets, and containerized applications
- Features include scalable, AI-ready storage with cyber resilience, reliable uptime, and simplified operations
UK Approves Green Belt Data Center Amid New Planning Policies
UK government approved a 96 MW Greystoke datacenter on Green Belt land near Abbots Langley, citing new NPPF policies prioritizing strategic infrastructure and site availability.
- UK government overruled Three Rivers District Council’s rejection of a 96 MW datacenter on Green Belt land near Abbots Langley, Hertfordshire, citing updated national planning policy.
- The revised National Planning Policy Framework (NPPF) now mandates local authorities to identify sites for datacenters, labs, gigafactories, and digital infrastructure, citing insufficient alternative sites.
- The proposed Greystoke datacenter spans up to 84,000 sqm (900,000 sq ft), includes ancillary facilities, and plans for a 25-pitch country park, with a £1.3 billion investment and £12 million for local education.
▶️ Open Source
Eric Migicovsky Chooses SiFli Microcontroller for Core Time 2 Upgrade
Eric Migicovsky selected SiFli’s SF32LB52J microcontroller with open source SDK, 512K SRAM, and low power profile for Core Time 2, replacing Nordic nRF52840 to support larger display and features.
- Selected SiFli SF32LB52J microcontroller chip for Core Time 2, featuring open source SDK, 512K SRAM, 16MB PSRAM, and a dedicated MIP peripheral
- SiFli chip offers low power consumption (~50μA with BLE connected), cost under $2, and supports 1-2MB SRAM options
- Previous microcontroller used was Nordic nRF52840; SiFli chip chosen due to open source SDK, higher RAM, and display interface needs
OpenAI Unveils Codex AI Coding Agent for Faster, Cleaner Software Development
OpenAI launched a research preview of Codex, an AI coding agent based on codex-1, capable of generating cleaner code, executing tasks in 1-30 minutes, and integrated with GitHub, initially for ChatGPT Pro, Enterprise, and Team.
- OpenAI announced a research preview of Codex, its most capable AI coding agent, on May 16, 2025
- Powered by codex-1, a version of the o3 AI reasoning model optimized for software engineering, producing cleaner code and adhering more precisely to instructions
- Runs in a sandboxed cloud environment, preloaded with user repositories via GitHub, completing tasks in 1-30 minutes, supporting multiple simultaneous tasks, with no browser or computer access restrictions
Aerospace Firm Exploits XOA Free Trials for a Decade, Creating 60+ Accounts
A $130M aerospace firm repeatedly exploited free trial limits of Xen Orchestra Appliance over 10 years, creating over 60 accounts via email manipulation, undermining open source sustainability.
- A $130 million aerospace company repeatedly exploited free trial periods of Xen Orchestra Appliance (XOA) over a decade, creating at least 60 accounts using corporate and personal emails.
- The company used incremental email addresses (e.g., johndoe01@outlook.com to johndoe60@outlook.com) to extend trial access without paying.
- The platform offers a fully open-source version and a supported appliance version, with the latter designed for easy deployment and updates, but the company avoided paying and support.
▶️ Software Development
Rust 1.87.0 Celebrates 10 Years with New Features and API Stabilizations
Rust 1.87.0, released on May 15, 2025, introduces anonymous pipes, safe architecture intrinsics, asm! label jumps, stabilized APIs, and deprecates the i586-pc-windows-msvc target, celebrating a decade of Rust development.
- Rust 1.87.0 released on May 15, 2025, marking the 10th anniversary of Rust 1.0
- Adds anonymous pipes to
std::io, enabling easier process stream management - Makes most
std::archintrinsics callable in safe code with target features enabled - Inline assembly (
asm!) now supports jumping to Rust labels for low-level control - Stabilizes
impl Traitprecise capturing in trait definitions - Stabilizes multiple APIs including
Vec::extract_if,split_off,extend_from_within, and variousdisplaymethods - Removes the
i586-pc-windows-msvctarget; users should migrate toi686-pc-windows-msvc - Includes numerous other API stabilizations and improvements
▶️ Management and Leadership
Coder Reverts to Manual Coding to Improve Quality and Debugging
After months of relying on LLMs for coding, the author reverted to manual coding to improve code quality, organization, and debugging, citing AI limitations and risk of losing programming skills.
- The author shifted from AI-assisted coding to manual coding with pen and paper after experiencing increased frustration and code inconsistency using LLMs like Claude and Gemini
- Initial AI use involved prompt-based code generation for infrastructure in Go and Clickhouse, but resulted in disorganized, inconsistent code resembling a junior developer’s work
- The author now reviews and rewrites code manually, limiting AI assistance to simple tasks, to regain control, improve debugging, and preserve coding skills
Stack Overflow Question Decline Signals Its Nearing Irrelevance
Stack Overflow’s question volume has fallen to 2009 levels due to moderation changes, pandemic effects, and ChatGPT’s rise, indicating the site is nearing irrelevance and may wind down.
- Stack Overflow question volume declined sharply since 2014, reaching levels comparable to its 2009 launch.
- Major decline correlates with improved moderation policies in 2014, pandemic-driven traffic surge in March 2020, and ChatGPT’s release in November 2022.
- Questions peaked during early pandemic lockdowns, then declined as ChatGPT provided similar or better answers, leading to near-zero monthly questions by May 2025.
Google’s 2011 Meeting Policies Were Bypassed by Teams to Skirt Limits
Google’s 2011 meeting policy reforms, including 50-minute hour-long meetings and calendar defaults, were circumvented by teams booking short slots, illustrating challenges in enforcing meeting discipline.
- In 2011, Larry Page became Google CEO and aimed to improve meeting efficiency with policies requiring decision-maker presence, 10-person cap, input from all attendees, and 50-minute hour-long meetings.
- Google Calendar was updated to default meeting durations to 25 and 50 minutes, but meetings often overran, ignoring these limits.
- Teams in Google’s NYC office exploited the policy by booking 50-minute meetings in the last 10 minutes of an hour, reserving the room for short standups, effectively bypassing the intended restrictions.
AWS CEO Urges UK to Expand Nuclear Power for Growing AI Data Centers
AWS CEO emphasizes Britain needs more nuclear power to support AI datacenter growth, with long lead times for nuclear projects and investments in SMRs, amid rising energy demands.
- AWS CEO Matt Garman states Britain requires increased nuclear power capacity to meet AI datacenter energy demands over the next ten years
- UK plans to invest £8 billion ($10.6 billion) in AI infrastructure in Britain by 2028, including multiple new datacenters and large-scale projects
- Global AI datacenter energy consumption is forecasted to more than double by 2030; UK energy demand for AI is expected to grow 500% in a decade
Atos’ ‘Genesis’ Plan Aims for €10B Revenue and 10% Margin by 2028
Atos launched the four-year “Genesis” plan to achieve €9-€10 billion revenue and 10% margin by 2028, focusing on AI, cost reduction, and regional restructuring amid financial and strategic challenges.
- Atos’ new four-year “Genesis” transformation plan aims for €9-€10 billion ($10-$11 billion) revenue and 10% operating margin by 2028
- Philippe Salle, the seventh CEO in three years, describes the company’s financial structure as “now secure”
- The plan includes job cuts, offshoring, AI-driven repositioning as a “global AI-powered technology partner,” and restructuring of business units, including Eviden and six regional hubs
78% of CISOs Face Rising AI-Driven Cyberattacks in 2025
Most CISOs (78%) already face AI-driven cyberattacks, with increased sophistication and automation; organizations seek simple, integrated AI security platforms like Darktrace’s Self-Learning AI to improve defenses amid personnel shortages.
- 78% of CISOs report experiencing AI-driven attacks already, according to Darktrace’s State of AI Cybersecurity 2025 survey of 1,500 IT security professionals
- 74% see AI threats as a current challenge; 90% expect ongoing impact; AI-fueled social engineering attacks increased 135% in 2023
- 95% believe AI enhances security speed and efficiency; 88% report significant time savings; 82% prefer AI solutions that do not share external data
Microsoft proposes concessions to address EU antitrust concerns over Teams bundling
Microsoft offers concessions including lower-priced Teams with Office suites, interoperability, and data portability for up to ten years to address EC antitrust concerns over bundling and market dominance.
- Microsoft proposes to offer Teams versions with Office 365 and Microsoft 365 at reduced prices, and allow customers to purchase suites without Teams for up to ten years.
- The company commits to enhanced interoperability with rival products, data portability, and the ability for clients to export Teams messaging data.
- The European Commission (EC) is reviewing these proposals, which include binding commitments for seven to ten years, to address antitrust concerns related to bundling and market dominance.
Microsoft Ends Free Nonprofit MS365 Business Premium Licenses
Microsoft is removing free MS365 Business Premium licenses for nonprofits, replacing them with discounted Business Basic plans and up to 300 licenses, impacting features and organizational costs.
- Microsoft is discontinuing free MS365 Business Premium licenses for nonprofits, replacing them with up to 300 licenses of Microsoft 365 Business Basic and discounts up to 75%
- The change was announced this week, surprising many users; services like Intune are absent from the new offerings, while Teams remains available
- Nonprofits must start paying for Business Premium after their current subscription ends, with significant feature reductions such as web-only Office apps and missing management services
NHS Trusts Question Palantir’s FDP Amid Limited Adoption and Concerns
English NHS trusts express concerns that Palantir’s FDP reduces existing system functionality; UK spends £330M on the platform, which faces limited adoption and skepticism over its benefits.
- Several NHS trusts in England report that Palantir’s Federated Data Platform (FDP) offers less functionality than existing systems, with only 45 of 215 trusts actively using it by end of 2024
- Leeds Teaching Hospitals Trust states adopting FDP would result in loss of current capabilities; they are exploring how their existing outpatient and discharge services compare
- NHS England claims FDP improves patient care coordination and hospital productivity, with over 120 trusts signed up and 72 trusts live in phased rollout; UK government allocated £330 million for the seven-year contract awarded to Palantir in November 2023
Trump warns Cook over manufacturing iThings in India due to tariffs and US plans
Trump warned Cook against manufacturing iThings in India, citing tariffs and US plans, amid India’s high import duties and India’s focus on diversifying its manufacturing sector.
- Donald Trump told Apple CEO Tim Cook he has a problem with manufacturing iThings in India.
- Trump expressed opposition to Apple building in India, citing tariffs and US manufacturing plans, despite Apple’s $500 billion US investment pledge.
- India imposes tariffs over 100% on agricultural imports and nearly 30% on manufactured goods; low wages attract manufacturers but high tariffs and policies influence production locations.
▶️ Technology
Ollama Launches New Engine Enhancing Multimodal Models and Vision Capabilities
Ollama’s new engine supports multimodal models such as Meta Llama 4 and Google Gemma 3, improving model modularity, inference accuracy, and memory management for advanced vision and multimodal tasks.
- Ollama introduces a new engine supporting multimodal models, starting with vision models like Meta Llama 4, Google Gemma 3, Qwen 2.5 VL, and Mistral Small 3.1
- The engine enhances model modularity, allowing each model to be self-contained with its own projection layer, improving reliability and ease of integration
- It improves inference accuracy by adding metadata for large images, managing memory with image caching, and optimizing KV cache and attention mechanisms for longer context sizes
9-Line Loop Boosts LLM Tool Integration for Developer Automation
A 9-line agent loop enables LLMs like Claude 3.7 to perform automated tasks with tools such as bash, enhancing developer workflows and automating complex, repetitive processes.
- Describes a simple 9-line loop for integrating an LLM with tool use in AI programming assistants like Sketch
- Utilizes Claude 3.7 Sonnet, enabling the LLM to perform tasks such as git operations, file editing, and command-line tool adaptation
- The loop handles tool calls, installs missing tools, and adapts to different command-line options, improving workflow automation
xAI Blames Unauthorized Prompt Change for Grok Bot’s Controversial South Africa Mention
xAI blames an “unauthorized modification” of Grok’s prompt for a bug causing it to repeatedly mention “white genocide in South Africa” on X, violating policies and following prior incidents.
- xAI attributed a bug in its Grok chatbot to an “unauthorized modification” of the system prompt on May 14, causing Grok to repeatedly reference “white genocide in South Africa” on X.
- The change, made around 3:15 AM PST, directed Grok to give specific responses on political topics, violating internal policies; this is the second incident linked to unauthorized modifications.
- Previously, in February, Grok was briefly instructed by a rogue employee to censor mentions of Trump and Musk; xAI plans to publish prompts publicly and implement stricter controls and monitoring.
Pioneering Personalized CRISPR Therapy Cures Baby with Rare Genetic Disorder
Doctors used a personalized CRISPR gene-editing infusion to cure a baby with CPS1 deficiency, pioneering a new era of tailored genetic therapies for rare disorders.
- KJ Muldoon, born with CPS1 deficiency affecting 1 in 1.3 million infants, received the first personalized gene-editing infusion designed specifically for his mutation.
- The treatment was administered at 9½ months old, marking the first case of a custom gene-editing therapy for a patient of any age.
- The approach demonstrates potential to treat thousands of rare genetic diseases through individualized CRISPR-based interventions.