Bitcoin Lightning node operators warned of AI-discovered security bugs

Bitcoin Lightning network node operators have received an emergency security warning following the discovery of critical bugs. The vulnerabilities were identified through AI-assisted work on Bitcoin code. Developers have chosen to withhold technical details for two weeks to allow operators sufficient time to apply fixes. This marks the second Lightning network security emergency this month linked to AI involvement in Bitcoin development.

12-Year-Old Developer Nears 1,000 Followers, Launches 7-Day AI Coding Event

A 12-year-old developer named Harun has amassed nearly 1,000 followers on Dev.to while building an AI mentorship tool called KODA on a budget Android phone. To mark the milestone, he has announced a seven-day global coding challenge open to developers worldwide. Participants are tasked with building a single-page website for a niche audience, with the requirement that at least half the code is generated using KODA. Winners will receive permanent in-app rewards including a crown theme, doubled memory, and a Pro Sensei Mode. Harun credited community members for helping him resolve a backend database error in his Supabase project during the event's lead-up.

Landsat 9 pass captured the effects of yesterday's catastrophic glacier collapse

Article URL: https://twitter.com/WxNB_/status/2092875025246261494 Comments URL: https://news.ycombinator.com/item?id=49463392 Points: 12 # Comments: 1

Dolly Parton, Jolene, and the power of women talking

A special session plan for delimitation Bills? Congress chief’s letter to PM

Developer discovers API bug that made users repeatedly confirm already-verified emails

A developer on DEV Community shared how a user spent an entire day re-confirming an already-verified email address because the API kept returning a misleading error message. The root cause was a single error response covering two unrelated conditions: an unverified email and a restricted preview API key. The user had copied a limited preview key into his workflow at signup, and verifying his email never upgraded that key's permissions. The fix involved both splitting the error messages to identify the correct problem and automatically upgrading preview keys the moment email verification is completed. The issue went undetected for months since the API returned a clean 403 response with no exceptions or alerts, and was only discovered after the user tracked down the developer on LinkedIn when the support email address turned out to be non-existent.

Why Archiving Full Prompt History Is Essential for Reliable AI Systems

When AI agents powered by large language models generate responses, the complete execution context — including user queries, retrieval results, system prompts, model settings, and tool calls — holds as much value as the answer itself. Unlike traditional software, LLM-based systems are non-deterministic, meaning the same input can yield different outputs if model weights, prompts, or retrieval data change. Prompt archival is the practice of storing this full execution trace for every AI interaction, enabling engineers to reproduce, debug, and evaluate system behavior over time. A standard trace should capture request metadata, input payloads, model configuration, completion outputs, tool invocations, and retrieval results at minimum. Without such records, diagnosing failures, answering stakeholder questions, or running meaningful A/B evaluations on AI pipelines becomes effectively impossible.

Air Conditioning Should Be Treated as a Basic Necessity, Experts Argue

A piece published on HumanProgress.org makes the case that air conditioning is no longer a luxury but an essential need for human wellbeing. The argument centers on the life-saving role AC plays in protecting people from extreme heat, which poses serious health risks. As global temperatures rise, access to cooling technology is increasingly seen as a matter of public health and safety. The article challenges the perception that air conditioning is an indulgence, positioning it instead as a critical infrastructure need.

Adentris (YC P25) Is Hiring

Article URL: https://www.ycombinator.com/companies/adentris/jobs/ZpMXZ0C-founding-engineer-ai-rcm-healthcare-platform-typescript-python Comments URL: https://news.ycombinator.com/item?id=49463361 Points: 0 # Comments: 0

Sleep score court is in session

Your honor, We all know that the woman who sits before you is tired. Just look at the circles under her eyes. The novelty Pride and Prejudice T-shirt and Muji sweatpants that pass for pajamas. The unkempt, banana clip hairstyle that says "I really did wake up like this." But she's doing her best. And that's why she deserves this court's time and attention, to correct the injustice of last night's sleep score.

Samsung Galaxy S26 FE launches with minor upgrades over its predecessor

Samsung has unveiled the Galaxy S26 FE, the latest entry in its Fan Edition smartphone lineup. The device retains the same 6.7-inch OLED display, 4,900mAh battery, and triple rear camera setup as last year's S25 FE. Notable changes are limited to a refreshed chipset and a slightly wider 12-megapixel front camera with an 85-degree field of view. Most of the new features are software-based, borrowed from Samsung's flagship phones released earlier this year. The phone is widely seen as an incremental update rather than a meaningful generational leap.

iCalendar's 75-Octet Line Limit Breaks Multilingual Feeds Using Multibyte Text

A developer building a multilingual iCalendar feed for AI model shutdown announcements discovered that the iCalendar spec (RFC 5545) enforces a 75-octet line limit, not a 75-character limit — a distinction that only surfaces when using multibyte character sets like Japanese UTF-8. While English text passed without issue, Japanese summaries silently violated the limit because each character can occupy up to 3 bytes, causing parsers to reject the feed. Fixing the issue required rewriting the line-folding logic to count bytes rather than characters, use proper CRLF line breaks, and handle multi-codepoint Unicode characters correctly. The developer also noted that calendar UIDs must be keyed on a locale-independent identifier, not localized text, to prevent duplicate events when users subscribe to multiple language feeds. A simple byte-length assertion run against a non-English test feed is recommended to catch this class of bug before deployment.

Two Laravel AI Eval Packages Compared: Pest Plugin vs Vizra Evals

Laravel developers testing AI agents now have two packages with overlapping names — pestphp/pest-plugin-evals and Vizra Evals — raising questions about which to use. Both packages share the same --evals flag and PEST_EVALS environment variable, meaning they can coexist in a single test suite without conflict. The Pest plugin offers a broad range of deterministic and model-scored expectations, including consistency testing via repeated prompt runs, making it well-suited for point-in-time quality checks. Vizra Evals, built on top of Pest, goes further by persisting every run to a database, enabling baseline comparisons and row-level regression detection across deployments. The key differentiator is whether a team needs to know if an agent performs well today, or whether it has degraded compared to a previous reference run.

Can NRIs invest in sovereign gold bonds?

Fermion Fleet Uses Structured Boolean Gates to Prevent AI Agent Overreach

Fermion Fleet is a multi-agent system developed for the Google All Things Agentic Hackathon that enforces order approvals through code-level boolean checks rather than relying on an AI model's confident-sounding language. The system requires a parseable structured approval before any order can be committed to a ledger, meaning ambiguous, malformed, or missing outputs automatically hold the transaction. To handle long-running conversations, the project employs a pressure-driven context eviction mechanism that moves lower-priority items to a recoverable pool instead of permanently deleting them, allowing earlier details to be recalled when later steps require them. The runtime is built on Google's Gemini 1.5 Flash via Vertex AI, Google ADK, and Cloud Run, while the permission and context policies are custom-built by the team. The current build stores context and ledger data in process memory, with persistence via services like Firestore noted as a planned but not yet implemented next step.

SendCheck tool validates crypto addresses in-browser before you hit send

A solo developer known as Pennyforge has launched SendCheck, a free browser-based tool designed to verify cryptocurrency addresses before funds are transferred. The tool checks address format, EIP-55 checksum validity, transaction history, and whether the destination is a wallet or smart contract. It also identifies USDC token variants and maintains a per-exchange, per-token, per-network support matrix for major platforms including Coinbase, Binance, and Kraken. All checks run client-side against public RPC endpoints, meaning no address data is sent to external servers. A free tier allows three checks per day, while unlimited checks cost a one-time fee of 0.50 USDC, verified on-chain.

Supreme Court cuts judge eligibility practice period from 3 years to 1

The Supreme Court has reduced the mandatory courtroom practice requirement for judicial service aspirants from three years to just one year. Under the revised framework, candidates will undergo two years of structured training as part of their preparation for judicial responsibilities. The change has triggered debate among legal experts about whether one year of practice provides sufficient courtroom experience. Former Justice Vipin Sanghi has argued in favour of retaining the original three-year practice period. However, lawyers Vivek Narayan Sharma and Siddharth Sijoria believe the new training-focused model can still produce competent judges.

Three-Layer Validation Pattern Cuts LLM JSON Pipeline Failures to Near Zero

LLM production pipelines frequently fail not due to flawed model logic but because model outputs violate JSON parsing contracts, causing error rates of 5–15% at scale. Engineers commonly rely on fragile regex hacks and try/except blocks, which cannot reliably handle issues like trailing commas, truncated strings, or plain-text safety refusals. A more robust approach uses a three-layer validation pattern: pre-sanitization of raw output, strict schema binding via Pydantic with provider-native structured outputs, and a lightweight repair fallback for malformed responses. OpenAI's structured outputs API, combined with Pydantic models, enforces token-level JSON compliance and delivers typed objects directly to downstream services. Developers are also advised to always check the finish_reason field for truncation and to avoid manual regex extraction entirely in favor of SDK-native parsing tools.

Gates Foundation, India to develop AI farm tools

Bengal Pilgrim Group of 33 Goes Missing in Nepal After One Member's Visa Denied

A group of 33 pilgrims from West Bengal has gone missing in Nepal, where they had traveled as part of a religious pilgrimage. One woman from the group was left behind after her Chinese visa was rejected, which inadvertently kept her safe. The rest of the group proceeded with the journey without her. Authorities are currently investigating the whereabouts of the missing pilgrims. The incident has raised concerns about the safety of pilgrimage groups traveling through the Himalayan region.