Today’s headlines highlight critical vulnerabilities in GitHub MCP risking private data exposure, NTT’s $10 billion investment in AI-ready infrastructure, Salesforce’s $8 billion acquisition of Informatica to boost AI capabilities, and the launch of Mistral’s new LLM-powered Agents API. Additionally, Meta restructures AI teams to accelerate product development amid industry shifts.
▶️ Internet Infrastructure
Critical GitHub MCP Vulnerability Exposes Private Data via Malicious Issues
Invariant revealed a critical vulnerability in GitHub MCP allowing attackers to exfiltrate private repository data via malicious issues, exploiting toxic agent flows detected by automated security analysis.
- Vulnerability affects the GitHub MCP server (github/github-mcp-server), with 14k stars, enabling attackers to access private repositories via malicious GitHub issues.
- Attack involves hijacking a user’s agent through a malicious issue, coercing it to leak private repository data into a public pull request.
- Invariant’s security scanner identified this as a toxic agent flow, a form of prompt injection that manipulates agent behavior without compromising MCP tools themselves.
NTT Global Data Centers Invests $10B by 2027 to Expand AI-Ready Infrastructure
NTT Global Data Centers announced over $10 billion investment by 2027, expanding capacity and deploying AI-optimized infrastructure, including liquid cooling, to support AI workloads and sustainability goals.
- NTT Global Data Centers commits over $10 billion in capital development through 2027 to expand infrastructure across North America, Europe, and Asia
- In 2024, delivered 370+ MW of new capacity, opened 10 new data centers, and supported over 200 MW of AI workloads, ranking as the third-largest global data center provider
- Invested $2.9 billion in land acquisitions and development, including a 68.5-acre campus in Johor Bahru, Malaysia (290 MW), and new sites in Berlin, Frankfurt, Chicago, and Hillsboro, enabling over 800 MW of future capacity
- Deployed advanced liquid cooling systems and designed facilities for high-density AI workloads to meet AI-driven infrastructure demands
- Achieved a 26% reduction in emissions and increased renewable energy use to 51% in FY2023, targeting net-zero emissions by 2030
AI Infra Summit 2025 in Santa Clara to Showcase AI Infrastructure Innovations
The 2025 AI Infra Summit in Santa Clara (September 9-11) features 200 speakers, 3,500 attendees, and four streams covering AI infrastructure, with key industry leaders sharing insights into AI-driven digital transformation.
- The 2025 AI Infra Summit will be held in Santa Clara from September 9-11, featuring four streams: AI Data Center, Edge AI, Enterprise AI, and Hardware and Systems
- Expected attendance is 3,500 with 35% enterprise end-users and tenfold increase in hyperscaler participation; around 200 speakers from major companies
- Keynote speakers include executives from Walmart, CVS Health, Boeing, Capital One, Lockheed Martin, Mastercard, PayPal, and Children’s National Hospital; over 120 partners will showcase technologies and solutions
Salesforce to Acquire Informatica for $8 Billion to Boost Data and AI Capabilities
Salesforce plans to buy Informatica for $8 billion to strengthen data management and AI tools, integrating metadata, catalog, and governance for a unified AI architecture.
- Salesforce to acquire Informatica for approximately $8 billion, announced on May 27, 2025
- Acquisition aims to enhance data management, governance, and AI capabilities, integrating Informatica’s data catalog, data integration, metadata management, and MDM with Salesforce platform
- Salesforce previously acquired Own Company for $1.9 billion in September 2024; total acquisitions over 10 years reach around $60 billion
▶️ Open Source
Hacker News Switches from Arc to SBCL for Better Performance and Scalability
Hacker News now runs on SBCL instead of Arc, using Clarc for performance and multi-core support, with code not yet open-source.
- Hacker News switched from Arc (built on Racket) to SBCL (Steel Bank Common Lisp) in September 2024 for performance improvements
- The change eliminates paging for long threads, enhancing user experience
- Clarc, an Arc-to-Common Lisp compiler, enables this transition and is in development, with the code not yet publicly released
- Clarc offers faster execution and multi-core support; its development has been ongoing since 2022
- Reimplementation involves staged Arc layers (arc0, arc1, arc2), with arc0 in the underlying system (Racket, JS, CL)
- Open-sourcing Clarc is feasible; releasing HN’s codebase is complicated due to anti-abuse measures and proprietary features
GitHub Issues Offers Free Unlimited Markdown Notes with Advanced Features
GitHub issues provides a free, unlimited note-taking platform with Markdown, inter-linking, search, and API support, but lacks offline sync; handles over 9,413 issues and 39,087 comments.
- GitHub issues offers free, unlimited notes with comprehensive Markdown support, syntax highlighting, and drag-and-drop media.
- Features include inter-linking issues across repositories, advanced search, and a comprehensive API for automation via GitHub Actions.
- Lacks synchronized offline support; privacy is considered secure, with recommendations to avoid storing sensitive data; supports checklists with
- [ ]syntax and issue referencing.
▶️ Software Development
How to Use Timeout with Bash ‘Until’ Loops Effectively
Using timeout with until in Bash requires wrapping until in a Bash process (timeout 1m bash -c "until ..."), or placing until in a separate script and applying timeout to that script.
timeoutcommand adds a time limit to commands, sendingSIGTERMby default after specified duration- Example:
timeout 1s sleep 5terminatessleepafter 1 second, returning exit code 124 - Direct use of
timeoutwith Bashuntilkeyword is not possible becauseuntilis a shell keyword, not a process
▶️ Management and Leadership
Introducing plwm: A Flexible Tiling Window Manager in Prolog
plwm is a customizable X11 tiling window manager in Prolog, supporting multiple layouts, multi-monitor, gaps, rules, hooks, external bars, and scriptability, with detailed configuration options.
- plwm is an X11 tiling window manager written in Prolog, powered by SWI-Prolog, with high configurability and modular design.
- Supports dynamic layouts (monocle, stack, grid, master-stack variants, nrows/N, ncols/N), multi-monitor, gaps, rules, hooks, animations, external bars, and scriptability via FIFO.
- Installation requires
xorgwithlibx11-dev,libxft-dev,libxrandr-dev, and SWI-Prolog; default install path is/usr/local/bin/.
Duolingo CEO Stands Firm on AI-First Strategy Amid User Backlash
Duolingo CEO Luis von Ahn’s May 2025 LinkedIn clarification failed to fully backtrack previous AI-first claims, maintaining plans to integrate AI deeply, risking user trust amid ongoing controversy.
- Duolingo CEO Luis von Ahn attempted to clarify his AI-first strategy amid user backlash, posting on LinkedIn on May 24, 2025.
- He acknowledged uncertainty about AI’s future impact but emphasized embracing new technology to stay ahead, without fully retracting previous statements.
- Von Ahn previously claimed AI was essential for scaling Duolingo and that headcount would only increase if automation was insufficient, with plans to replace contractors with AI.
Salesforce to Acquire Informatica for $8 Billion in Early 2027 Deal
Salesforce will acquire Informatica for $8 billion, paying $25 per share, in a deal expected to close early in fiscal year 2027, funded by cash and debt.
- Salesforce Inc. agreed to acquire Informatica Inc. for approximately $8 billion, with shareholders set to receive $25 per share, a 30% premium over prior close.
- The deal follows a 59% share decline for Informatica since last year’s initial talks failed.
- Salesforce plans to fund the acquisition through a combination of cash and new debt, with expected closing early in fiscal year 2027.
Google CEO Pichai Praises Jony Ive’s Legacy Amid OpenAI’s $6.5B AI Hardware Deal
Google CEO Sundar Pichai praised Jony Ive’s design legacy and highlighted AI’s transformative potential, amid OpenAI’s $6.5 billion acquisition of Ive’s startup IO to develop AI hardware products.
- OpenAI acquired Jony Ive’s startup IO for nearly $6.5 billion to develop AI hardware products, collaborating with his design collective LoveFrom.
- Google CEO Sundar Pichai praised Ive’s design legacy, including work on iMac, iPhone, and Apple Watch, and highlighted AI’s potential to surpass the internet in impact.
- Pichai emphasized ongoing Google AI innovation, expecting a “family of AI products,” and noted the timing of OpenAI’s announcement coincided with Google I/O 2025.
CISA Leadership Exodus in May 2025 Sparks Concerns Over Cybersecurity Stability
Most of CISA’s top officials have left or will leave in May 2025 due to government downsizing, risking reduced cybersecurity leadership during escalating threats.
- Most top officials at CISA, including leaders of 5 of 6 operational divisions and 6 of 10 regional offices, are leaving or have left in May 2025 amid government downsizing.
- Key departures include acting heads of divisions (e.g., Steve Harris, Trent Frazier, Vince Delaurentis) and regional directors (e.g., John Durkin, Jay Gamble, Alex Joves), impacting agency operations.
- Administrative leaders such as Val Cofield, Tarek Abboushi, Juan Arratia, and Blair Duncan are also departing, raising concerns over agency expertise and strategic capacity.
Reclaiming Education: Embracing Analog Methods to Combat AI Cheating
ADH advocates for restricting AI use in education through analog methods, citing AI’s role in cheating, superficial learning, and mental health issues, inspired by the Butlerian Jihad metaphor.
- ADH discusses the cultural and pedagogical challenges posed by AI, emphasizing the rise of anti-AI sentiment and the metaphor of the Butlerian Jihad from Dune.
- Highlights the increasing use of AI for cheating in education, with examples of students and teachers struggling with AI-generated work, and the limitations of current detection methods.
- Proposes returning to analog methods like handwritten work and participatory grading to foster genuine learning and reduce AI dependency in classrooms.
Apple Satellite Project Faces Resistance and Competition from SpaceX
Apple’s satellite ambitions are threatened by internal resistance and Elon Musk’s SpaceX competition, with delays and strategic disputes impacting plans for global satellite connectivity involving thousands of satellites.
- Apple’s satellite project faces internal resistance and competition from Elon Musk’s SpaceX
- The effort aims to provide global satellite connectivity, potentially involving thousands of satellites
- The project’s progress is hindered by internal disagreements and strategic challenges within Apple
Striving for Simplicity in Linux: Why Minimal Distros and BSDs Lead the Way
The article examines the persistent difficulty of maintaining Unix/Linux simplicity, favoring minimal distros like Alpine Linux and BSDs such as FreeBSD for their streamlined, less bloated design.
- The article discusses the ongoing challenge of achieving simplicity in Unix/Linux operating systems, emphasizing the difficulty of adhering to the KISS principle.
- It highlights the preference for smaller, less bloated distros like MX Linux and Alpine Linux, which are considered more manageable and less irritating.
- The article references various Linux distributions and BSD variants, including OpenBSD, FreeBSD, Chimera Linux, Void Linux, and eweOS, analyzing their features and limitations related to system complexity and functionality.
Nick Clegg Warns Copyright Rules Could Kill UK AI Industry
Nick Clegg argues that copyright consent laws would end the UK AI industry; UK Parliament favors protections, but government blocks mandatory disclosure, risking legal and ethical conflicts.
- Ex-Meta executive Nick Clegg claims that requiring AI companies to obtain copyright holder consent before training models would “kill the AI industry overnight”
- UK House of Lords voted in favor of amendments to the Data (Use and Access) Bill to protect copyrighted works from AI copying; government ministers blocked the amendment
- Clegg, former UK deputy PM and Meta president of global policy, expressed skepticism about the feasibility of asking creators for permission prior to training AI models, citing data volume challenges
- UK government promotes AI development through initiatives like “AI Growth Zones” and a licensing model allowing use of copyrighted data with payments
- Many UK media and arts professionals oppose data access bills; AI companies, including OpenAI, have mined copyrighted content without explicit permission
- UK Copyright Office head was dismissed after ruling that AI training on copyrighted material exceeds fair use; UK and US policies increasingly favor AI training exceptions
- Clegg’s stance aligns with government and big tech interests, contrasting with industry critics and creative sector advocates advocating for creator rights and transparency
US iPhone Manufacturing Challenges and Impact of Tariffs
Despite threats of tariffs, iPhone manufacturing in the US is unlikely due to high costs and lengthy setup, requiring years and significant investment, making tariffs ineffective.
- Morgan Stanley analysts estimate a US-made iPhone would be at least 35% more expensive than overseas models due to tariffs and higher US labor costs
- A 25% tariff on Chinese or Indian iPhone imports would increase global prices by only 4-6%, maintaining profit margins
- Apple would need 2-4 years to establish US manufacturing facilities, train over 100,000 skilled workers, and build multiple factories, with timelines similar to TSMC’s Arizona chip fab
UK HMRC to Invest £500 Million in Modernized Call Center Services
UK HMRC will spend £500 million on a CCaaS contract to modernize call center services, addressing poor performance and high legacy system costs, with implementation until December 2033.
- UK HM Revenue and Customs plans to award a £500 million ($677 million) contract for contact center as a service (CCaaS) to replace legacy systems, expected in December 2025.
- The contract includes designing, implementing, supporting, maintaining, and optimizing call center services until December 2033, with possible extension to December 2035.
- The current system’s poor performance in 2023-24 resulted in 44,000 customers cut off after 70-minute waits, with an average wait time of over 23 minutes; total on-hold time in 2023-24 was equivalent to 798 years.
- HMRC’s digital services, including Making Tax Digital, have increased costs by £300 million; legacy system remediation costs reached £482 million in 2023-24, with major programs extending scope and costs by 60-390%.
▶️ Technology
Lossless Video Compression Using Rational Bloom Filters and Probabilistic Hashing
The project proposes lossless video compression via Rational Bloom Filters, leveraging probabilistic hash counts and sparse frame differences for efficient, exact reconstruction.
- Introduces lossless video compression using Rational Bloom Filters, implemented in youtube_bloom_compress.py
- Applies probabilistic non-integer hash function counts (k*) to optimize Bloom filter parameters based on input data density
- Compression is effective when the 1s density in binary data is below approximately 0.32453, with parameters derived from input size and density
Mistral Launches Agents API with Multi-Function LLM-Powered Autonomous Systems
Mistral launched the Agents API on May 27, 2025, offering LLM-powered autonomous systems with features like server-side conversation management, code execution, image generation, web search, document libraries, and agent handoffs, mirroring recent industry-wide advancements.
- Mistral announced a new “Agents API” on May 27, 2025, enabling autonomous systems powered by large language models (LLMs) to plan, use tools, and achieve goals.
- The API manages conversation state server-side, allowing message threading without local history maintenance.
- Features include code execution via Code Interpreter, image generation with Black Forest Lab FLUX1.1 [pro], web search with two versions (
web_searchandweb_search_premium), document library for user-uploaded documents, and Model Context Protocol support. - Supports agent handoffs for task delegation among multiple agents, enabling complex multi-agent workflows.
- The rollout of these features aligns with similar updates from OpenAI (May 21), Anthropic (May 22), within eight days.
Meta Restructures AI Teams to Boost Product Development and Innovation
Meta reorganizes its AI teams into AI Products and AGI Foundations units to accelerate product rollout, focusing on Llama models, reasoning, multimedia, and voice, amid competitive pressures.
- Meta restructures AI teams into AI Products and AGI Foundations units to accelerate product development and enhance capabilities, including Llama models and reasoning, multimedia, voice.
- The AI Products team manages Meta AI assistant, AI Studio, and features within Facebook, Instagram, and WhatsApp; AGI Foundations focuses on reasoning, multimedia, voice, and Llama models.
- No layoffs or executive departures are reported; the reorganization aims to improve flexibility and speed amid talent attrition to rivals like Mistral.
Computex 2025 Highlights: Quirky Custom Tech and Gaming Innovations
Computex 2025 showcased innovative and quirky tech, such as customizable earbuds, themed keyboards, Jensen Huang merchandise, and advanced cooling and gaming furniture, highlighting emerging trends and niche markets.
- Taiwan’s Computex 2025 featured unconventional tech products, including customizable earbuds with acrylic nail extension slots and themed mechanical keyboards resembling buns.
- Biosong’s B7 earbuds allow users to paint or have nail salons decorate acrylic panels, with LED-lit designs and optional charms, targeting accessory market expansion.
- Nvidia CEO Jensen Huang’s keynote drew large crowds, with merchandise including a NT$1100 ($37) Jensen Huang party shirt; other highlights included liquid-cooled PC waterfall mods and ASUS’s ergonomic gaming furniture and Thunderbolt 5 GPU docking station.
Reducing AI Energy Consumption Through Innovation and Open-Source Solutions
Reducing AI’s massive energy use requires lateral thinking, open-source innovation, and economic incentives, as industry-driven race for larger models ignores environmental and efficiency concerns.
- AI’s energy consumption is difficult to estimate but is driven by an arms race among big tech companies prioritizing faster token generation and larger models.
- Current industry practices focus on performance and feature expansion without regard for environmental impact, with costs serving as barriers to entry.
- Solutions proposed include adopting power-efficient techniques, open-source ecosystems, and controlling data to reduce reliance on energy-intensive AI training and inference.
Sundar Pichai Predicts AI Will Surpass the Internet and Transform Google
Sundar Pichai states AI will surpass the internet in significance, emphasizing a new platform shift with real-time, multimodal AI search, hardware innovations, and potential for broad product and business impact.
- Sundar Pichai predicts AI will be “bigger than the internet” and discusses its profound impact on platforms, search, and hardware.
- Google’s AI products, including Gemini, Veo 3, Imagen, and AI Mode, represent a new phase of platform shift emphasizing self-improving, multimodal AI.
- Google’s AI Mode builds custom, interactive search results in real time, expanding traffic to more websites despite opposition from publishers like the News Media Alliance.