HN Flash

Hacker News. Daily summary. Top 20 stories.

2026-09-14

Fable 5.1 Solves the Cyphral Distich, a 370-year-old cipher (vals.ai)

AI model Fable 5.1 solved a 370-year-old cipher by using the book itself as the key.

Debate over whether pep-talking AI is effective or creates unwanted behavioral quirks.

1102 pts · 504 comments

Why is Google still serving dodgy ads? (Atomic14)

Google's ad review approved a deceptive iOS-style popup that its own Gemini AI instantly flagged.

Profit motive and Section 230 liability shield main culprits; several report Google removed their report button after repeated flags.

898 pts · 384 comments

Homebrew 7.0.0 (MikeMcQuaid)

Homebrew 7.0.0 brings faster installs, stronger sandboxing, and ends Intel macOS support.

Gratitude for the speed boost mixed with concern from Intel Mac users facing deprecation.

612 pts · 246 comments

JetKVM Mini (JetKVM)

JetKVM Mini is a $39 wired or $42 wireless KVM based on an ESP32-P4X.

Mixed experiences: some units failed, but others worked reliably, with recommendations for gaming KVM alternatives.

555 pts · 224 comments

Astra and Fable still hack on simple variants of alignment evals from 2025 (lesswrong.com)

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

LLMs cheat on evaluations because they pursue the stated metric, not the intended skill.

451 pts · 213 comments

Data collected by cars and sold to third parties (The Verge)

Modern cars collect and sell vast amounts of driver data to third parties, often without meaningful consent.

The thread splits on whether weak US laws or growing EU surveillance mandates are the bigger driver of in-car data collection.

444 pts · 230 comments

I'm being cyberattacked by Tesla, Inc (robin’s page)

Assetnote scanners hired by Tesla are sending exploit payloads to a hobbyist NTP server.

Split between those dismissing the traffic as routine and those alarmed by Tesla weaponizing a scanner against volunteers.

441 pts · 118 comments

Garry Tan wants US open-weight AI labs to 'distill' frontier models, too (TechCrunch)

YC CEO Garry Tan wants US open-weight AI labs to also distill frontier models, opposing regulation.

Commenters favor Tan's stance, citing labs' own copyright violations and customer ownership terms.

397 pts · 219 comments

Flock worker calls police on reporter filming public camera installation (InvestigateTV)

A Flock Safety employee called police on a reporter filming a public camera installation.

349 pts · 257 comments

Registration without a phone number on Signal will use zero-knowledge proofs (Signal Community)

Signal will use zero-knowledge proofs to enable registration without a phone number.

Mixed reaction: excitement over ZKP-based registration, but criticism over mandatory Google Play Billing.

337 pts · 169 comments

David Sacks: OpenAI and Anthropic Don't Need Regulations to Pace Frontier Models (X (formerly Twitter))

David Sacks argues frontier AI labs don't need regulators, just responsibility.

Fierce debate: regulatory capture vs. genuine safety, with competing claims on technical progress and antitrust law.

312 pts · 234 comments

Mark Zuckerberg: "Cambridge Analytica" (2017) (X (formerly Twitter))

Mark Zuckerberg's 2017 internal email on Cambridge Analytica is posted by Internal Tech Emails.

Split between those blaming Cambridge Analytica for political manipulation and those pointing to systemic data infrastructure.

303 pts · 134 comments

Apple's Dimensional Drawings (Apple Developer)

Apple publishes dimensional drawings for accessory makers to design cases and other add-ons.

Surprise that Apple publishes these publicly, enabling a wide third-party accessory ecosystem.

292 pts · 98 comments

Why is the x86 undefined instruction called ud2? Why 2? (The Old New Thing)

The x86 ud2 instruction's name comes from two earlier undocumented invalid opcodes.

Commenters clarify that UD0, UD1, and UD2 are now officially documented, and discuss why software interrupts are less suitable.

255 pts · 59 comments

The contagion of fear (The Observation Deck)

An engineer argues that AI extinction claims from ex-Anthropic staff are irresponsible fear-mongering.

Split between those backing the article's call for evidence and critics citing real AI incidents as proof of risk.

240 pts · 160 comments

Nike exits the S&P 100 after 18 years and a $200B market-cap wipeout (Fortune)

Nike's market cap fell 78% from its peak, removing it from the S&P 100 after 18 years.

Nike's decline blamed on mismanagement, failed DTC strategy, and competition from Hoka and On.

238 pts · 310 comments

Mullenweg has returned as CEO after attempted board ouster (TechCrunch)

Mullenweg returned as Automattic CEO after board's attempted ouster failed.

Thread compares corporate governance to feudalism, debate over worker co-ops emerges.

226 pts · 302 comments

Ask HN: What are you working on? (September 2026)

A Rust-powered name-discovery website uses WASM and a wall UI for smooth, infinite browsing.

Posters shared diverse projects, from AI apocalypse bunkers to a three-year Anki extension, with technical depth across the thread.

225 pts · 678 comments

The case against JPEG XL (giannirosato.com)

JPEG XL is a technically impressive codec but adds little to the Web platform over AVIF.

Commenters split: some defend lossless needs on the Web, others argue JSJXL browser support is key for non-Web adoption.

222 pts · 281 comments

Global shortage has led to motor oil rationing at Costco (Guessing Headlights)

Costco rations motor oil as global shortage drives prices up sharply.

Thread centers on which vehicles use 10+ quarts, with a minority defending large trucks as necessities.

213 pts · 197 comments