Meituan’s LongCat team released MIT-licensed weights for LongCat-2.5, a multimodal agent-oriented model designed for long, multi-step tasks such as coding. The
「Background」Huawei Ascend accelerators are a domestic alternative to Nvidia GPUs for large-scale AI training. Earlier Horizon reporting said DeepSeek is helping shape future Huawei AI chips to reduce reliance on Nvidia, while another report said Huawei customers must source their own HBM for the next-generation Ascend 960DT, underscoring supply constraints in China’s advanced AI hardware stack.
「Impact」Developers and organizations building agentic coding tools can evaluate LongCat-2.5 as a commercially usable, MIT-licensed model with a reported 1M-token context window and API pricing from $0.300 per million input tokens and $1.20 per million output tokens, plus a two-week free OpenCode trial. Because the Nvidia-free training claim and Terminal Bench result are vendor-reported, teams should benchmark the model on their own workloads and validate runtime compatibility before replacing proprietary coding models.
A blog post arguing that all AI-generated code should be open source sparked a substantial Hacker News discussion about software licensing, provenance, and copy
「Background」The proposal rests on two related legal and licensing questions: whether code produced with substantial AI assistance can be treated as human-authored copyrighted software, and whether models trained on open-source code create obligations for the output. In this context, “open source” means releasing the source under a license that permits inspection, modification, and redistribution, while CC0 is a public-domain dedication that removes most copyright restrictions.
「Community Discussion」Commenters disagreed over the legal and practical implications: one AGPL author claimed early LLMs reproduced their code and modern systems still borrow its design, while others cited U.S. Copyright Office guidance and cases such as Doe v. GitHub and Thaler v. Perlmutter to argue that AI output is not automatically open source and may not be copyrightable without human authorship.
A discussion of OLED MacBook Pro displays centers on pulse-width modulation flicker as a potential usability problem for sensitive users. The comments cite low-
「Background」PWM flicker is the rapid pulsing used to control display brightness, and sensitive users can report headaches, dizziness, and eye strain depending on the modulation frequency. The MacBook Pro OLED is Apple’s tandem OLED PROLED XDR display, introduced alongside the Mini-LED MacBook Pro configurations that have been available since 2021.
「Impact」For users sensitive to display flicker, the practical consequence is that a MacBook Pro OLED upgrade may still cause eye strain, headaches, or concentration problems if PWM flicker is present at usable brightness levels. The supplied context includes a PWM flicker test on a tandem OLED MacBook Pro but does not report its result, and notes that independent measurements of the specific OLED panel were not yet available, so sensitive users should verify flicker behavior at their normal brightness before relying on the device for long work sessions.
「Community Discussion」Commenters split between technical skepticism and personal sensitivity: one reader questions the 240 Hz claim by proposing out-of-phase tandem OLED layers, while another says 240 Hz would make hand-waving look stop-motion. Other users describe experiences with flicker, CRI, motion blur, or post-concussion screen discomfort, but these are anecdotal rather than evidence of the MacBook Pro’s behavior.
The supplied report says Nvidia’s security chief quietly left in August, and CEO Jensen Huang has since stepped down over his relationships, including one with
「Background」Earlier Horizon coverage reported that Nvidia announced a CEO transition involving Jensen Huang without confirming a successor, effective date, or cause. It also said Malaysia and Singapore escalated investigations into suspected Nvidia AI chip smuggling, including charges tied to 917 servers and an inventory audit by a company central to the controversy to account for restricted GPUs.
SpaceX’s Starship Flight 15 ended with the upper stage being caught, according to the linked SpaceX launch page and announcement. The supplied primary source is
「Community Discussion」Commenters framed the catch as a step toward fully reusable launch, with one comparing it to the Shuttle’s expensive refurbishment and another noting that heat-shield durability and final payload mass fraction remain the practical tests. A reposted summary claimed a roughly 22.5-hour, 14-orbit flight, 26 Starlink V3 satellites deployed, and a booster splashdown, but those details were not verified by the supplied primary source.
Research reports that apps built with Lovable exposed passport details and more than 200,000 private messages. Lovable’s head of security says the root cause wa
「Background」The report concerns apps built with Lovable on Supabase, where row-level security (RLS) is the mechanism that limits what browser clients can read from a backend. Horizon's September 26 digest earlier reported that some Supabase customers were publicly exposing user data, underscoring risks in AI-generated and vibe-coded applications. That earlier coverage provides context for why Supabase RLS defaults and Lovable's security controls are central to the current incident.
「Impact」Developers using Lovable apps backed by Supabase should audit Row-Level Security and Data API exposure, because missing or overly permissive policies can allow visitors to query private tables with the public anon key. Supabase’s October 30, 2026 default change reduces exposure for new public tables, but apps that relied on implicit table access may need explicit grants, and existing projects may still require remediation. If sensitive data was exposed, affected app owners should also consider rotating compromised credentials or keys.
「Community Discussion」Commenters debate responsibility, with some arguing that frontend-to-Supabase row-level security is a minefield and that Supabase’s onboarding defaults contributed, while Lovable’s security lead emphasizes insufficient RLS and automated scanning or fixing. One commenter frames the incident as a warning that platforms need deny-by-default databases, least-privilege APIs, and stricter publishing controls.
PyPI has redesigned project pages to surface clearer security and provenance signals for packages. The update is live and is intended to help Python package use
「Background」PyPI's provenance system uses attestations to record where a package file came from and how it was built or published. Horizon's October 21 digest reported an alleged compromise of LiteLLM 1.107.0 on PyPI that delivered a credential stealer and backdoor, illustrating why clearer trust signals on project pages matter.
「Impact」The immediate consequence for PyPI users is that provenance status becomes more visible on project pages, allowing them to check whether release files have attestations before installing a package. However, the presence of provenance is not an absolute security guarantee: if an attacker controls the CI identity, a malicious release can still produce valid Sigstore or SLSA attestations. Teams should therefore treat provenance as a supply-chain signal to inspect, not as standalone approval.
「Community Discussion」Commenters welcomed the greater provenance visibility but debated its trust model: some argued that trusted publishing and sigstore-style keyless signing could centralize trust around major CI/CD platforms, while others noted that the signals mainly expose compromised PyPI entries rather than inherently malicious projects.
Cloudflare has open-sourced Kitesurf under Apache 2.0, a stateless browser for AI agents built on Cloudflare Workers. The project parses HTML, handles CSS, and
「Impact」Developers building AI-agent workflows now have an experimental, Apache 2.0-licensed browser engine they can inspect, run locally, or deploy on Cloudflare Workers, with Browser Run integration for screenshots, HTML extraction, and automation. Because Kitesurf is stateless and Workers-based rather than a conventional Chromium container, teams should verify whether their existing agent tooling and web-automation dependencies remain compatible before replacing current browser infrastructure.
AI score (0–10) × 10 × source weight × time decay. Source weight: official first-party ×1.2, established media ×1.1, community discussion ×1.0, aggregators ×0.9. Time decay uses a 24-hour half-life, so older stories sink naturally instead of camping on the list. Scores are produced by an LLM rating content value, independent of any commercial relationship.