Today’s top stories include Google’s expansion of sovereign cloud services to host US Top Secret data, Brazilian researchers scraping over 2 billion Discord messages raising privacy concerns, and Nvidia’s CEO criticizing US-China AI chip export restrictions as a policy failure fueling Chinese innovation.
▶️ Internet Infrastructure
Brazilian researchers scrape 2.05B Discord messages for research amid privacy concerns
Brazilian researchers scraped 2.05 billion public Discord messages from 3,167 servers (2015-2024) via Discord’s API, publishing a 118GB compressed database for research, raising privacy concerns despite anonymization.
- Brazilian researchers scraped over 2.05 billion Discord messages from 3,167 servers (2015-2024) using Discord’s public API, covering approximately 10% of servers.
- Data was anonymized, with usernames replaced by pseudonyms, IDs hashed and truncated, and no personally identifiable information retained.
- The dataset, published as JSON files, totals 6.2GB uncompressed (108GB expanded), intended for research on mental health, politics, misinformation, moderation, and chatbot development.
Google Expands Sovereign Cloud Services for US Top Secret Data Hosting
Google has enhanced its sovereign cloud services, including air-gapped and region-specific platforms, to address geopolitical concerns and enable US government Top Secret data hosting, supporting global data sovereignty.
- Google has expanded its sovereign cloud offerings, including an air-gapped platform for high-security needs, authorized to host US Top Secret and Secret data.
- The Google Cloud Air-Gapped solution is fully standalone, does not require external network connectivity, and is aimed at defense and intelligence sectors.
- Google Cloud Dedicated, developed with Thales and operated by S3NS, supports AI workloads with GPU services, currently in preview, and plans to expand globally, starting with Germany.
- Google Cloud Data Boundary enables customers to control data storage and processing locations, with features like US/EU content processing limits and the new User Data Shield leveraging Mandiant cybersecurity services.
- Google has formed regional partnerships in Europe, Asia, and the Middle East to deliver sovereignty-compliant cloud services, amid geopolitical tensions and US data sovereignty concerns.
- The company’s initiatives follow similar moves by AWS and Microsoft, which have launched European-specific cloud regions and data boundaries to address local sovereignty requirements.
- Google emphasizes a flexible, layered approach to sovereignty, offering solutions tailored to regulatory and risk profiles, amid increased global demand for data localization and security.
Greater Manchester NHS Analytics System Outperforms UK-Funded Platform
Greater Manchester NHS claims its analytics system is 2-3 years ahead of the UK government’s Federated Data Platform, citing superior capabilities and concerns over FDP’s actual adoption and scope.
- Greater Manchester ICB reports its NHS analytics platform surpasses the capabilities of the UK government-backed Federated Data Platform (FDP), which is built under a £330 million (approximately $442 million) seven-year contract awarded in November 2023.
- The local Data, Insight and Intelligence Team estimates their system is 2-3 years ahead of FDP in functionality, despite NHS England’s push for rapid FDP adoption across trusts, with over 120 trusts signed up and 72 already live.
- The report highlights concerns about FDP’s actual usage, noting many trusts onboarded do not actively use it, and emphasizes that FDP is not a like-for-like replacement for Greater Manchester’s existing Analytics and Data Science Platform (ADSP), which offers broader capabilities.
Active Exploitation of Ivanti EPMM Cloud Vulnerabilities and Malware Attacks
Ongoing exploitation of CVE-2025-4427 and CVE-2025-4428 in Ivanti EPMM has reached cloud instances, with attackers deploying malware like Sliver; patches were issued last week, but active attacks persist.
- Exploitation of CVE-2025-4427 (authenticated bypass) and CVE-2025-4428 (post-auth RCE) has extended to cloud instances of Ivanti Endpoint Manager Mobile (EPMM)
- Vulnerabilities affect on-premises EPMM, with Wiz confirming active exploitation in cloud environments since May 16, 2025
- Attackers deployed remote-control malware Sliver, using IP 77.221.157.154, linked to previous exploits in Palo Alto Networks appliances; US CISA added both CVEs to its Known Exploited Vulnerabilities Catalog
Microsoft backs Epic’s fight against Apple’s App Store restrictions to launch Xbox store
Microsoft supports Epic Games’ legal challenge against Apple’s App Store restrictions, aiming to launch its Xbox mobile store and enable off-app purchases, but Apple’s policies and appeals threaten deployment.
- Microsoft filed support for Epic Games’ legal fight against Apple’s App Store policies on May 21, 2025.
- The support addresses Apple’s anti-steering policies and restrictions on linked-out payments, which hinder Microsoft’s Xbox mobile store launch.
- The court ruling permits Microsoft to launch its Xbox mobile store with off-app purchase links, but Apple’s appeal could delay or revoke this access.
▶️ Open Source
Google Unveils Gemma 3n, a Mobile Multimodal AI with Fast On-Device Performance
Google introduces Gemma 3n, a mobile-optimized, multimodal AI model with 5B/8B parameters, reduced RAM usage via PLE, and real-time on-device performance for diverse applications.
- Gemma 3n is an open model previewed on May 20, 2025, built on a new architecture optimized for mobile, developed with Qualcomm, MediaTek, and Samsung.
- Features include 5B and 8B parameters with Per-Layer Embeddings (PLE), enabling models to operate with a 2GB or 3GB memory footprint, respectively.
- Capabilities encompass faster on-device response times (1.5x faster), multimodal understanding (audio, text, images, video), high-quality speech recognition and translation, and improved multilingual performance (e.g., 50.1% on WMT24++ ChrF).
Litestream 2.0 Boosts SQLite Recovery with LTX and Scalable Replication
Litestream has been revamped with LTX-based point-in-time restores, conditional write support for generation management, and a VFS read-replica layer, enhancing scalability and performance for SQLite applications.
- Litestream is an open-source tool enabling reliable recovery of SQLite databases from object storage, supporting full-stack applications.
- Major update introduces LTX-based point-in-time recovery with fast restore via transaction-aware compaction, merging multiple changesets.
- Implements CASAAS: conditional write support in object stores like S3 to manage WAL stream generations, allowing ephemeral nodes and overlapping instances.
- Adds lightweight read replicas using a VFS layer that caches pages from S3-compatible storage, improving performance for hot reads.
- Supports replication of large numbers of databases via directory globbing, overcoming previous scalability limitations.
- Last updated on May 20, 2025, emphasizing ongoing development and industry relevance.
▶️ Software Development
Microsoft’s GitHub Copilot AI sparks criticism over low-quality PRs and overhyped claims
Microsoft’s GitHub Copilot agent entered public preview, generating low-quality PRs on .NET code, sparking community criticism over AI’s limited reliability and overhyped expectations.
- Microsoft announced the public preview of GitHub Copilot coding agent on May 19, 2025
- Deployed to open PRs on the .NET runtime repository, with many PRs containing low-quality or erroneous code
- Community reactions highlight AI-generated PRs as often unhelpful, error-prone, and indicative of overhyped AI capabilities
AI Cannot Replace Developers Without Mastering Office Politics and Team Dynamics
AI cannot replace developers until it understands office politics, teamwork, and context; coding is a small part of the job, with human intuition essential for innovation and security.
- Article argues AI cannot replace developers until it comprehends office politics and team dynamics
- Writing code constitutes a small part of a developer’s role; most work involves collaboration, planning, and contextual decision-making
- Coding often involves novel, uncharted territory requiring human intuition, trial-and-error, and gut instinct, especially in areas AI cannot assist with
- The proliferation of connected devices has increased the complexity and importance of reliable, secure software, demanding skilled human oversight
- AI tools trained on existing resources cannot fully understand or navigate the nuanced, human aspects of software development
▶️ Management and Leadership
Meta Boosts “Below Expectations” Ratings to 20% Amid Workforce Restructuring
Meta is expanding its “Below Expectations” ratings to 15-20% in mid-year reviews, aligning with efforts to streamline workforce and accelerate performance-based layoffs, following prior adjustments in 2022.
- Meta instructs managers to increase employees rated “Below Expectations” to 15-20% in mid-year reviews, up from 12-15% last year
- The review process, starting June 16, allows performance-based exit decisions, including employees who resigned or were dismissed for underperformance
- The move follows Meta’s previous effort in late 2022 to double the share of employees in lowest performance categories, amid ongoing workforce restructuring
TikTok US E-Commerce Layoffs Amid Restructuring and Slowing Sales
TikTok announced US e-commerce layoffs on May 21, with staff told to work remotely during organizational restructuring aimed at improving efficiency amid declining sales caused by tariffs and economic factors.
- TikTok Shop US employees were instructed to work from home on May 21 ahead of organizational and personnel changes.
- The layoffs are part of a broader effort to create more efficient operating models amid declining US sales due to global tariffs and economic uncertainty.
- The company did not disclose the number of affected employees; changes follow a previous restructuring of TikTok’s e-commerce governance in April.
Nvidia CEO Calls US China AI Chip Restrictions a Failure
Nvidia CEO Jensen Huang condemned US chip export restrictions on China as “a failure,” citing their role in boosting Chinese AI hardware development and prompting US policy reversal to support domestic innovation.
- Nvidia CEO Jensen Huang called US export controls on China’s AI chips “a failure,” citing accelerated Chinese development and market share decline from 95% to 50% over four years.
- Huang stated export restrictions have fueled Chinese rivals, including Huawei, to develop competitive AI hardware using local technology and second-tier chips.
- The US ban on Nvidia’s product tailored for China led to a $5.5bn writedown; Nvidia has no plans to release further “Hopper” series chips for China.
- Huang criticized Biden-era AI diffusion rules, arguing they hinder US leadership; the Trump administration is reversing these regulations to promote US tech growth.
- US government warnings against using Chinese AI silicon, including Huawei’s Ascend processors, have prompted Beijing to push domestic AI chip purchases to reduce reliance on imports.
Nvidia CEO Criticizes US Export Bans on China as Harmful and Counterproductive
Jensen Huang condemned US GPU export bans to China, stating they undermine Nvidia’s growth and global AI progress, while China’s hardware development suggests policy failure.
- Nvidia CEO Jensen Huang criticized US export bans on high-performance GPUs to China as “precisely the wrong policy” and “a failure”
- Huang argued the bans harm Nvidia’s revenue, US tax contributions, and global AI research, citing China as home to half of the world’s top AI researchers
- He highlighted that Chinese firms are developing their own AI hardware, indicating US policies may spur local innovation instead of slowing it
▶️ Technology
Google Search Introduces AI Mode and Flow to Revolutionize Search and Filmmaking
Google Search is integrating AI Mode for conversational, follow-up queries, and launched Flow, an AI-driven film tool, to enhance search and content creation amid AI industry competition.
- Google Search will incorporate AI Mode, launching in a new tab to enable conversational queries and follow-up questions.
- The update was announced at Google I/O, emphasizing Search’s evolution amid rising AI competition from OpenAI and Anthropic.
- Google introduced Flow, an AI-powered moviemaking tool aimed at studios, potentially reducing high production costs and disrupting traditional filmmaking.
Best Foldable Phones of 2025: Book-Style Designs Lead the Way
In 2025, the best foldable phones are primarily book-style, with the Google Pixel 9 Pro Fold leading due to its large display, premium build, and versatile features, while alternatives like the OnePlus Open and Samsung Galaxy Z Flip 6 offer high performance and innovative designs.
- The article reviews the best foldable phones in 2025, emphasizing book-style, tablet-hybrid designs over clamshell flip phones.
- Top models include the Google Pixel 9 Pro Fold, OnePlus Open, Samsung Galaxy Z Flip 6, and Motorola Razr (2023), with detailed specifications and performance assessments.
- The Pixel 9 Pro Fold features an 8-inch tablet display, 6.3-inch cover display, Tensor G4 processor, IPX8 rating, and advanced camera system, making it the overall best.
HDR Technology and Formats Transform TV and Gaming Visuals
HDR enhances contrast, brightness, and color range on TVs, supported by formats like Dolby Vision and HDR10+; premium displays reach 1,500–4,000 nits, improving realism and detail.
- HDR enhances contrast, brightness, and expanded colors, with brightness measured in nits; premium models reach 1,500–4,000 nits, entry-level around 400–700 nits
- HDR is often combined with Wide Color Gamut (WCG), supporting color spaces like Rec. 2020 and DCI-P3 for a broader color spectrum
- There are four main HDR formats: HDR10, HDR10+, Dolby Vision, and HLG; HDR10 is standard, while HDR10+ and Dolby Vision support dynamic scene-by-scene metadata for improved tone mapping
- Dolby Vision and HDR10+ offer more accurate HDR images than HDR10, with industry support favoring Dolby Vision; major streaming services support both
- HDR gaming is supported on PS4, PS4 Pro, PS5 (HDR10 only), Xbox One X/S, Xbox Series X/S (HDR10 and Dolby Vision), with PC support for HDR10, HDR10+, and Dolby Vision; Nintendo Switch currently lacks HDR support but will support HDR10 in its upcoming model
Google I/O 2025 Unveils AI as a Co-Worker with New Features and Integrations
Google I/O 2025 revealed a transition to AI-driven tools that enhance user productivity, with new models like Gemini 2.5, AI Mode in Search, and integrations across Chrome, Gmail, and Meet, signaling a major shift in AI utility and web interaction.
- Google I/O 2025 announced a shift from AI passive reading to active user collaboration, emphasizing AI as a coworker and superpower.
- New AI features include AI Mode in Search, Gemini 2.5 with enhanced reasoning, Project Mariner web-browsing agent, and tools like NotebookLM Video Overviews and Google SynthID AI Content Detector.
- Google plans to integrate Gemini AI assistant into Chrome, Gmail, and Google Meet, with upcoming features like live translation, smart replies, and AI-generated videos; stock dropped 1.5% post-announcement.
Hacker Breaches US Officials’ Messages, Exposes Metadata Threatening National Security
A hacker accessed over 60 US government officials’ messages via TeleMessage, exposing metadata that could threaten counterintelligence, with the platform suspended since May 5 amid security concerns.
- Hacker breached communications app used by former Trump national security adviser Mike Waltz, intercepting messages from over 60 US government officials
- Intercepted data includes messages from disaster responders, customs officials, diplomatic staff, White House personnel, and Secret Service members, covering a day ending May 4
- Metadata from the breach poses counterintelligence risks, with experts highlighting potential intelligence access despite limited content sensitivity
AI’s Growing Energy Footprint and Environmental Impact
MIT Technology Review estimates AI’s energy footprint from single queries to national data centers, projecting exponential growth in energy demand and emissions, driven by industry investments and limited transparency.
- MIT Technology Review analyzed AI energy use, estimating single query consumption at approximately 57 to 3,353 joules depending on model size.
- AI data centers in the US used around 200 TWh in 2024, with projections indicating AI-specific energy demand could reach 165–326 TWh annually by 2028, powering 22% of US households.
- Major tech companies plan to invest over $500 billion in new data centers and nuclear power to support AI growth, with limited transparency on actual energy consumption and emissions.
AMD Unveils Threadripper Pro 9000 Series and RDNA 4 GPU with Cutting-Edge Specs
AMD’s Threadripper Pro 9000 series, based on Zen 5, offers up to 96 cores and 128 PCIe 5.0 lanes, with RDNA 4 GPU featuring 32GB vRAM, debuting in July.
- AMD announced its Threadripper Pro 9000 series with up to 96 cores, 128 PCIe 5.0 lanes, DDR5 6400 MT/s ECC memory, and boost clocks up to 5.4GHz
- The non-Pro Threadripper 9000 variants offer up to 64 cores, 128 threads, and similar clock speeds, with TDP of 350W
- AMD’s RDNA 4 architecture debuts in the Radeon AI Pro R9700 GPU with 32GB vRAM, 96 TFLOPS FP16, and 1531 TOPS INT4 performance; release expected in July
Google Unveils AI Coding Agent Jules and Expands Gemini 2.5 Integration
Google announced public beta of AI coding agent Jules, integrated Gemini 2.5 models into development tools, and showcased new AI features across Chrome, Android Studio, and Firebase, expanding AI capabilities and models.
- Google announced public beta of AI coding agent Jules, capable of autonomous code generation, bug fixing, and testing on GitHub repositories.
- Google’s Gemini 2.5 Pro model integrated into Android Studio introduces app testing (Journeys) and dependency updates (Version Upgrade Agent).
- Google IO showcased Stitch, an AI service for UI design generation supporting export to CSS/HTML or Figma, with models Gemini 2.5 Pro and Gemini 2.5 Flash supporting real-time multilingual speech.
- Chrome 138 includes built-in Gemini Nano AI features, Summarizer, Language Detector, Translator, and Prompt APIs; Firebase received AI enhancements including Figma import.
- Google launched AI Ultra subscription at $250/month with Gemini 2.5 Pro, 30 TB storage, and YouTube Premium; Gemini models are expanding to Chrome on Windows/macOS, with real-time query capabilities via Gemini Live.
- Gemini Code Assist is now generally available, and Google Colab is integrating Gemini 2.5 Flash for AI-assisted data science workflows.
- Open-source Gemma models, including Gemma 3n and MedGemma, have new releases; upcoming models include SignGemm and DolphinGemma for sign language and dolphin communication understanding.