Rare English moth rediscovered at Northumberland estate after 70-year absence

A rare moth species, missing from England for 70 years, has been rediscovered at the National Trust's Wallington Estate in Northumberland. Conservationists made the find in just 10 minutes, crediting extensive habitat restoration work at the site. The moth had previously been considered at risk of extinction in England. The rediscovery has raised hopes that other hidden or presumed-lost insect populations may also be found through similar conservation efforts.

Trump Claims US Does Not Need Canada as Trade Tensions Escalate

US President Donald Trump asserted that Canada needs the United States more than the other way around, amid an intensifying trade dispute between the two nations. The statement comes as trade tensions between Washington and Ottawa continue to rise. Canada ranks as the second-largest trading partner of the United States, following Mexico. Energy trade forms a central pillar of the bilateral economic relationship between the two countries.

Google Gemini Spark Offers Persistent Background Task Automation Across Devices

Google has introduced Gemini Spark, an agentic layer within the Gemini platform designed to execute multi-step tasks in the background without requiring the user to stay active. Unlike Gemini Live, which focuses on voice-first conversations, Spark is built to take a defined task and continue running it even after an app or device is closed. A key demonstrated capability involves starting a complex workflow on a mobile device and having Spark complete it on a macOS host computer. The feature supports conversational task setup, cross-device orchestration, and integration with Google Workspace. As of June 2026, Spark is available only to Google AI Ultra subscribers aged 18 and over in the United States, with a macOS app beta also underway.

XRP Ledger Order Book Dominates AMM with 89% of Trading Volume

Despite the activation of the Automated Market Maker (AMM) amendment on the XRP Ledger in 2024, the platform's traditional order book continues to handle the vast majority of trading activity, accounting for 89% of 24-hour volume. Recent data shows the order book processed approximately 7.35 million XRP across over 108,000 trades, while AMM pools handled around 912,000 XRP across 42,000 trades. Analysts attribute the order book's dominance to the XRPL's low transaction costs and fast settlement, which make professional market-making economically viable on-chain — unlike most other blockchains where high gas fees push liquidity into passive AMM pools. The XRPL has supported an on-chain central limit order book since 2012, giving it a structural advantage that newer smart-contract chains lack. This dynamic challenges the broader industry assumption that AMMs have become the dominant liquidity venue across all blockchain ecosystems.

Meta agrees to sweeping teen safety changes on Instagram and Facebook

Meta has reached a child safety settlement with attorneys general from 51 US states and territories, requiring major changes to how teens use Instagram and Facebook. The agreement mandates new safeguards and usage restrictions for minors on both platforms. The settlement arises from a nationwide lawsuit accusing Meta, Google, TikTok, and Snap of failing to protect children and deliberately designing addictive features. Under the deal, Meta must implement controls that limit when and how teenagers can interact with its social media services. The agreement marks one of the most significant regulatory actions targeting social media companies over child safety concerns in the US.

ONNX Runtime Web Now Enables Full AI Inference, Background Removal Inside Browser

Developers can now run complex deep learning models, including background removal and feature extraction, directly in the browser using ONNX Runtime Web (onnxruntime-web). This eliminates the traditional need to send user data to remote servers, addressing latency, bandwidth, and privacy concerns tied to regulations like GDPR and HIPAA. The approach relies on three pillars: the ONNX model format for cross-framework portability, the onnxruntime-web runtime compiled via WebAssembly, and browser-native compute APIs such as WebGL and WebGPU. Models originally trained in PyTorch, TensorFlow, or JAX can be serialized into ONNX and executed locally on the user's device CPU or GPU. The shift effectively turns the browser into an edge-computing node capable of real-time computer vision tasks without any cloud infrastructure.

How India turn the screw against Sri Lanka

Meta Pays $18B to Settle Child Safety Claims With Most US States

Meta has reached an $18 billion settlement with nearly all US states over claims that its platforms posed risks to children's safety. As part of the deal, Meta will introduce daily usage limits on its social media platforms for minors. The settlement represents one of the largest child-safety-related payouts by a tech company. However, Florida rejected the agreement, with state officials dismissing the sum as insufficient given the scale of harm alleged. The deal marks a significant legal and regulatory milestone in ongoing efforts to protect children online.

How to Build a Lean, Region-Aware Uptime Stack for SaaS MVPs

A practical guide for SaaS teams outlines a minimal but effective observability stack combining external uptime probes, self-hosted health endpoints, structured logs, and scheduled-job heartbeats. The approach is designed for products serving both EU and US customers, where data residency and retention controls are critical considerations. Rather than maximising telemetry volume, the focus is on answering specific incident questions — such as whether a failed nightly pipeline affected the public API or only internal imports. Key recommended metrics include job duration, records processed versus rejected, and queue depth labelled by region, while high-cardinality data like customer emails must be kept out of metric labels. The guide advises teams to map failure modes before selecting tools, and to choose self-hosted infrastructure when compliance requirements outweigh setup convenience.

New Tool Lets Servers Deliver Markdown to AI Agents via Accept Headers

A new web tool called AcceptMarkdown proposes a method for servers to serve Markdown-formatted content specifically to AI agents. The approach leverages HTTP Accept headers, allowing clients to signal they prefer Markdown over HTML. This could streamline how AI systems consume web content by receiving cleaner, more structured text. The project was shared on Hacker News, where it garnered modest early attention with 12 points and one comment.

Google Expands Gemini With Daily Brief, Spark Agent and Deeper Workspace Integration

Google has announced a major update to its Gemini app, introducing three key features: Daily Brief, Gemini Spark, and an enhanced Gemini Live experience. Daily Brief generates a personalized morning digest by pulling from a user's Gmail, Calendar, and Gemini chats, highlighting priorities and suggested next steps. Gemini Spark is a planned 24/7 AI agent designed to complete tasks on a user's behalf across Workspace tools like Gmail, Docs, and Slides, even after the app is closed. Gemini Live is now fully integrated into the main Gemini experience, allowing users to switch seamlessly between text and extended voice conversations. Daily Brief will initially roll out to Google AI Plus, Pro, and Ultra subscribers in the United States, while Gemini Spark will begin with trusted testers before a beta release for Ultra subscribers.

Nepal Flash Floods: Over 100 Dead, 133 Indians Among 750+ Missing

Devastating flash floods originating in Tibet have swept through central Nepal, killing more than 100 people. Over 750 individuals remain missing in the aftermath, including at least 133 Indian nationals. Among those unaccounted for is a 32-member group from Kolkata, along with several other travelers from West Bengal. The floods have caused widespread destruction across central Nepal, triggering urgent search and rescue operations.

Bill Gates: AI Risks Are Real but Can Be Managed

Bill Gates published an essay in 2023 arguing that while artificial intelligence poses genuine risks, those risks are manageable. Writing on his personal blog GatesNotes, he acknowledged concerns around AI safety and misuse but maintained an overall optimistic outlook. Gates emphasized the importance of ensuring AI benefits are accessible to everyone, not just a privileged few. He called for thoughtful governance and collaboration to guide AI development responsibly.

New Tool Tracks GitHub Outages in Real Time

A lightweight web tool called 'Is GitHub Cooked' has been launched to monitor GitHub's service availability. The site provides real-time status updates on whether GitHub is experiencing an outage. It appears to be an unofficial, community-built alternative to GitHub's own status page. The tool was shared on Hacker News, where it garnered modest early attention.

Vibe Coding: Five security risks when developers over-rely on AI tools

A growing software development trend called 'Vibe Coding' involves delegating nearly all programming logic to AI tools like GitHub Copilot or ChatGPT, with developers acting as high-level directors rather than hands-on coders. While the approach boosts productivity, security experts warn it is normalizing dangerous vulnerabilities in production environments. Common critical errors include accidentally pushing secret credentials to public repositories, storing authentication tokens in insecure browser storage, leaving admin routes unprotected, skipping server-side input validation, and exposing sensitive data through debug logs. AI models are designed to produce working code quickly but lack inherent security judgment, meaning unchecked output can introduce serious exploitable flaws. Experts stress that the developer remains ultimately responsible for auditing AI-generated code and ensuring it meets industry security standards.

Case-Sensitive Linux Filesystem Exposed a Bug That Windows Silently Ignored for Months

A developer's first Vercel deployment failed because a CSS file was imported as 'Randomfacts.css' while the actual filename was 'randomfacts.css' with a lowercase 'r'. The error had gone undetected locally for months because Windows filesystems are case-insensitive and resolved both spellings to the same file without complaint. Vercel builds on Linux, which treats uppercase and lowercase filenames as entirely distinct, causing the build to fail with an unresolved import error. The fix required matching the import to the real filename and, due to a Git quirk on Windows, renaming files with a case-only change requires a two-step process using a temporary filename. The developer's key takeaway was to deploy early in a project's lifecycle, since deployment environments like Linux can expose an entire category of bugs that a Windows development machine will never surface.

Buzz Framework Examined: Engineering Tradeoffs in Agentic AI Workspaces

Buzz is an event-driven workspace designed for human-agent collaboration, built around shared identity, auditable actions, and a signed event history. A technical analysis published on DEV Community evaluates the system across throughput, latency, capacity planning, backpressure, and operational cost. The article argues that collapsing requests, observations, decisions, and effects into a single untyped message makes Buzz easy to demonstrate but difficult to audit at scale. Typed message boundaries and provenance-tracked tool calls are presented as essential safeguards against unauthorized actions being mistaken for completed tasks. The authors conclude that Buzz functions less as a standalone tool and more as a protocol requiring durable records and evidence-backed outputs to remain reliable under pressure.

Omarchy Linux Workstation Design Prioritizes Reliability Over Feature Visibility

Omarchy is an opinionated Linux workstation built on Arch Linux and Hyprland that treats defaults, applications, navigation, updates, recovery, and documentation as a single unified product. A technical guide published on DEV Community outlines an engineering model for implementing Omarchy with a focus on state ownership, failure handling, and measurable outcomes rather than surface-level features. The guide emphasizes structuring the system lifecycle into distinct phases — admission, validation, execution, commitment, observation, and recovery — each with a stable identifier and a clear owner. It argues that technology choices like shell scripting and migrations are implementation materials, not architectural conclusions, and only become meaningful when tied to constraints and failure modes. The core design principle is that every shortcut must preserve a traceable path back to explanation, enabling the system to fail loudly, recover deliberately, and improve from real-world evidence.

Florida Catholic Bishops Reject Religious Exemptions Pushed by State AG

Florida's Catholic bishops have declined to support religious exemptions from vaccine requirements, effectively rejecting a push by the state's Attorney General. The bishops' decision runs counter to the stance of Florida's political leadership, which has been broadly critical of vaccine mandates. By refusing to provide religious cover for exemption claims, the Church distanced itself from the state's anti-vaccine posture. The move is notable given Florida's ongoing efforts to challenge vaccine policies at both the state and federal level.

Developer Runs Three Projects Using Seven Claude Code AI Subagents as a Virtual Company

A developer has built a structured multi-agent system using Claude Code, assigning seven distinct roles — including CEO, researcher, writer, and marketer — to guide decisions across three small projects. Each role is defined by a written instruction file specifying its responsibilities, decision-making authority, and escalation rules. The workflow follows a fixed sequence, from status reporting through research, drafting, design, and marketing review, before a final CEO-level sign-off. Despite the system's sophistication, all three projects — a Japanese blog, a local shop listing service, and an English build-log site — have earned no revenue as of the post's publication in August 2026. The author acknowledges the system is a work in progress and published this post as a transparent account of the process rather than a success story.