HN Flash

Hacker News. Daily summary. Top 20 stories.

2026-09-19

Android 17 is the first since 3.x to add new APIs without releasing to the AOSP (GrapheneOS Mastodon)

Google is withholding standard Android APIs and security patches from non-Pixel devices, GrapheneOS says.

Strong frustration with Google's Android API gatekeeping; some see a fork as inevitable but unlikely soon.

917 pts · 501 comments

Microsoft exec called AI scraping 'the largest theft of labor in human history' (TechCrunch)

Unsealed lawsuit filings reveal Microsoft exec called AI scraping 'largest theft of labor in human history.'

Widespread agreement that the scraping is theft, with debate over whether regulation would make things better or worse.

900 pts · 796 comments

I don't like passkeys (Ethan Hawksley)

Passkeys offer strong anti-phishing security but introduce real risks of permanent account lockout for individuals.

Commenters split: passkeys are fine for synced-ecosystem users but dangerous for anyone who needs cross-device or cross-platform recovery.

795 pts · 771 comments

Cloudflare Quick Tunnels (try.cloudflare.com)

Cloudflare now offers instant public URLs for local servers without requiring an account.

Mixed reactions: appreciation for the ease of use, but concerns about latency variance and trusting Cloudflare with cleartext traffic.

756 pts · 296 comments

Claude Code now reads AGENTS.md if there is no Claude.md (Claude Code Docs)

Claude Code now reads AGENTS.md as fallback when no CLAUDE.md exists.

Praise mixed with frustration: finally supporting the standard, but still lacking .agents/skills.

684 pts · 249 comments

OpenJev (openjev.com)

OpenJev runs a local decision model in the browser using GGUF-quantized LLMs.

The thread focuses on whether OpenJev's approach is novel or just a facsimile of Jev.

649 pts · 273 comments

AI-generated posters don’t have to be horrible (‘ERE I AM - JH!)

A designer shows how specific style prompts can make AI event posters less generic.

Two camps: those who see real improvement with specific prompts, and those who still detect an unmistakable AI uniformity.

480 pts · 302 comments

A heap overflow and SSO misconfiguration to compromise OpenAI internal repos (Hacktron AI)

Researchers chained a libheif heap overflow and OpenAI SSO flaw to access internal repos.

Fury over the $6,500 bounty for a chain that could leak OpenAI's source code, with debate on black-market value.

479 pts · 203 comments

US Military had close call after using AI for hallucinated intelligence report (cnn.com)

A US military analyst's use of AI generated a false intelligence report that nearly led to an attack on a Chinese ship.

Two camps: one blaming AI itself, the other arguing the military already had an accountability problem that AI just compounds.

471 pts · 353 comments

Human brain is two separate organs, Stanford Medicine-led research finds (News Center)

Forebrain and hindbrain arise from separate progenitor cells, not a single lineage.

The thread splits between genuine interest in the developmental finding and criticism of the headline's claim of two organs.

397 pts · 148 comments

Saving another 100TB of RAM (Cloudflare Blog)

Cloudflare reclaimed 100TB+ of RAM by optimizing consistent hashing and its Rust structs.

Readers split on whether siloed abstractions are necessary or overblown, with praise for the math-driven optimization.

394 pts · 86 comments

Jemalloc 5.4.0 (GitHub)

Article body wasn't reachable. HN discussion still summarized.

Split on jemalloc's front-page worthiness, with strong testimony on its practical value for runtimes and memory management.

331 pts · 94 comments

Bend 2 and the Vibe-Coding Trap (blog.liampwll.com)

Vibe-coded formal verification language Bend 2 ignores existing tools like SPARK.

Sharp split: some see a credible project, others suspect hype and fake engagement.

320 pts · 232 comments

Inside ZCode: Silently uploading your Git history to the cloud (Code is cheap, let's talk)

ZCode silently uploads entire workspaces and Git history to Aliyun OSS, encrypted with a server-only key.

Widespread condemnation of silent exfiltration; comparisons to Grok Code scandal.

315 pts · 107 comments

Korea raises data breach fines to 10% of revenue (koreajoongangdaily)

South Korea raises maximum data breach fines to 10% of annual revenue.

Skepticism over whether chaebols will actually face the 10% fine, versus praise for the deterrent effect.

314 pts · 105 comments

Warren Buffett Steps Down as Berkshire Chairman, Names Son to Replace Him (nytimes.com)

Warren Buffett steps down as Berkshire Hathaway chairman, appointing his son Howard.

Nepotism criticism dominates, with some noting the chairman role is symbolic and culture-focused.

308 pts · 203 comments

The scourge of x86 emulation (fex-emu.com)

Emulating x86's strict memory model on ARM causes severe performance penalties without hardware TSO support.

Valve's FEX emulation draws debate over bytecode vs. emulation for ARM game support.

281 pts · 93 comments

ZCode, the GLM coding agent, silently uploads your Git history (tokenstead.ai)

ZCode silently uploads entire Git history encrypted to Alibaba Cloud, users cannot decrypt.

Top comments split: some criticized the article's wording, others noted similar practices elsewhere.

260 pts · 14 comments

San Francisco Onion Futures Company (onionfutures.com)

A site sells private, transferable onion futures contracts, claiming legal loophole.

Two camps: performance art vs. legal gray area; law originated from 1950s market cornering.

251 pts · 88 comments

I vibed a proof of Conway's conjecture (overreacted.io)

A math noob used AI to produce a Lean proof of Conway's 50-year-old refinement conjecture.

Confusion over the surreal number construction dominates, with helpful clarifications offered.

249 pts · 239 comments