October 2026
166 editions archived
Saturday, October 31
230 threadsA backdoored Claude Code extension with 27 million installs sat on the Cursor Marketplace, and that was only the start of the day. Binance lost $6.3B in 20 minutes, Northflank and…
- Breach after breach: the trusted tooling layer is failing bearish
- Cloudflare as the web's gatekeeper, and the bot arms race mixed
- AI writes the code, humans are told to write the spec mixed
- Effort is no longer a quality filter, and readers are angry bearish
- Infra startups face a flight to hyperscalers, but hyperscalers leak too bearish
- Federated and open alternatives keep getting built bullish
- Building small and for joy, as a reaction to burnout bullish
Friday, October 30
230 threadsThe official Claude Code extension on Open VSX, and through it Cursor's marketplace, shipped a backdoor, and one commenter says he confirmed Binance uses that extension. The same …
- Dev tooling is the new attack surface bearish
- Open weights keep closing the gap, with licensing strings mixed
- Platform risk keeps biting: Anthropic, Cloudflare, AWS bearish
- AI-sounding text is now a trust problem bearish
- Agent tooling is hardening around terminals, sims and validation bullish
- Tiny models doing surprising things: Doom and a Pico bullish
- Political fight over an 'AI bubble bailout' starts early bearish
- Low-level craft and infra hiring still draws real interest neutral
Thursday, October 29
230 threadsGitLab upgraded to 19.3.6 and the hole was still open, because the real fix lived in a Ruby patch that only ships in GitLab's official packages. The bug was an integer overflow in…
- Vulnerabilities are piling up and everyone blames the agents bearish
- Frontier model race continues, with a focus on cheating mixed
- Agent tooling has hidden costs and strange behavior mixed
- Adversarial text: poisoning, magic strings, and fake papers mixed
- Stack Overflow under 1,000 questions a month bearish
- Hardware and energy get cheaper and more automated bullish
- Craft and distrust of vibe-coded infrastructure mixed
- DoorDash buys Instacart, and consolidation shrugs bearish
Wednesday, October 28
230 threadsTen days after the eMudhra certificate authority was found compromised, Git on macOS still trusts it, and the thread is mostly people arguing that the real failure is Apple shippi…
- Trust infrastructure keeps failing in boring, unfixed ways bearish
- AI is moving into verifiable, high-stakes work mixed
- Local AI gets a hardware push while cloud coding gets squeezed mixed
- Human-made is becoming a feature people brag about mixed
- Platform dependency keeps biting small builders bearish
- Hardware is getting a speed and sovereignty pitch bullish
- Government opacity and rules aimed at tech mixed
- Compensation data and the hiring grind stay opaque bearish
Tuesday, October 27
230 threadsAnthropic's S-1 hit the front page with no discussion attached, and a preprint on OpenAI's GitHub claiming the Exponential Time Hypothesis is false sat right beside it. The labs l…
- Frontier labs go public and claim real science mixed
- AI made work faster, not lighter mixed
- Crawlers, slop, and the fight for human-only spaces bearish
- Platform dependence keeps biting builders bearish
- AI inference security: leaks, scrapers, and protocol sprawl mixed
- Open source maintainers are out of bandwidth mixed
- Independent web businesses face an uncertain road mixed
- Solar passes nuclear, but US install costs lag bullish
Monday, October 26
230 threadsSomeone reran a zstd Rust rewrite benchmark on Anthropic models and found that even Haiku 5.5 can pull it off, and that one comment sits at the center of a day where the question …
- Model prices collapse as open weights go on-prem mixed
- Be more ambitious: AI as a lever, not a replacement bullish
- Slop anxiety: trust in human-made work is eroding bearish
- Small, specialized models and context discipline beat brute force mixed
- Web bloat backlash: static, fast, printable neutral
- Infrastructure fragility: Cloudflare blocks, AWS billing bug bearish
- Hardware is still hard, and builders love it anyway neutral
Sunday, October 25
231 threadsA commenter says OpenAI banned a researcher's account for reverse engineering a Samsung TV, and the thread's reaction was swift: if you can't point an LLM at a black box, it loses…
- Closed-model access risk pushes builders toward open and local mixed
- AI coding splits builders: bigger ambition, or hollowed-out craft mixed
- Proof of human work is becoming a product problem bearish
- Creative jobs shrink, and AI gets blamed for part of it bearish
- Infrastructure keeps breaking in boring, expensive ways bearish
- Inference hardware: squeezing GPUs, wafer-scale racks, and narrow models mixed
- A craft revival: small tools, hard constraints, hand-built software bullish
Saturday, October 24
231 threadsA hacker pointed an LLM at a Samsung TV, found an on-board AI listening in, and got his OpenAI account suspended. In the same 24 hours an iPhone hack claim landed, a new HTTP2Shel…
- A brutal week for security, and everyone feels it bearish
- AI vendors cut you off, and open models are the escape hatch mixed
- AI slop is now a verification problem, not a taste problem bearish
- Who owns the device: governments, Apple, banks, and TV makers bearish
- Python tooling is consolidating under an AI lab mixed
- Performance craft is back in fashion bullish
- For graphics, LLMs work better writing code than making pixels bullish
- Delightful small tools built on simple data bullish
Friday, October 23
230 threadsQuad9 is leaving France, eMudhra is quitting the SSL certificate business, and Docker pulls in Spain keep dying whenever La Liga plays. Three different places, one lesson: the neu…
- Neutral infrastructure is being squeezed out by liability bearish
- AI bug-hunting is compressing the patch window bearish
- AI vendors tighten control over who can use them bearish
- Agent infrastructure gets real: sandboxes, runtimes, and covert channels mixed
- Vibe coding hits the trust wall mixed
- Why doesn't OpenAI just run 100 companies? The answers are telling neutral
- Privacy fights shift to payments, TVs and phones mixed
- Builders are still making things just because bullish
Thursday, October 22
230 threadsOn Tuesday around 15:00 UTC, two PHP CVE IDs started circulating on social media with the wrong release time and a wrong CVSS score of 10.0, hours before the scheduled fix. Amazon…
- Coordinated disclosure is breaking down in public bearish
- Supply chain and platform secrets keep getting stolen bearish
- AI labs trade cloud deals while tightening their grip mixed
- The trust gap with AI agents and LLM judgment mixed
- Code is clay: AI as a power tool for experts bullish
- DuckDB 2.0 LTS pulls the data stack toward one process bullish
- Platforms keep rewriting what 'owning' a game means bearish
- Meta's $25B New Mexico penalty reopens the tort reform fight mixed
Wednesday, October 21
230 threadsThe Dutch National Cyber Security Center's public website went offline today because it runs PHP, and a critical PHP security release is about to land with no affected versions an…
- PHP critical CVE sends governments offline before the patch bearish
- Web PKI breaks again: eMudhra CA keys cracked in milliseconds bearish
- Open-weight DeepSeek gets a serving stack; US labs' moat gets questioned bullish
- Coding-agent vendors under fire: effort levels, downgrades, and usage caps mixed
- AI slop is turning into a trust tax on content and communities bearish
- Privacy and trust leaks: Meta's Muse and a ransom negotiator turned extortionist bearish
- Rented-land risk: platforms break things your business depends on bearish
- AI as a second wind for older and disabled builders bullish
Tuesday, October 20
230 threadsDeepSeek open-sourced the inference engine behind its V4.1 Flash service on the same day DHS banned it from federal civilian agencies. One side shipped a cheap, fast, inspectable …
- DeepSeek open-sources its engine as Washington moves to ban it mixed
- Local and open models are now 'enough' for daily coding bullish
- Shipping AI code nobody understands, and who vouches for it mixed
- Hidden instructions and baked-in behavior erode trust in AI bearish
- Concentrated infrastructure keeps failing in public bearish
- Agent benchmarks and harness design get serious bullish
- Reclaiming your own devices and stack bullish
Monday, October 19
230 threadsA satirical pitch for 'completely hand-made' software, every line crafted by a real human, led today's front page, while down the list people were installing LiteLLM proxies and t…
- The agent harness is the new battleground bullish
- Rate limits and subscriptions push people toward owning their stack mixed
- Human authorship is becoming a feature, and a fight mixed
- New model releases get 'where are the benchmarks?' treatment mixed
- Shared infrastructure keeps surprising people bearish
- Flock, ICE, and who gets to look people up bearish
- Hardware and low-level tinkering is having a moment bullish
Sunday, October 18
230 threadsDeepSeek shipped v4.1-Flash, and the replies were mostly disbelief at the benchmarks plus a quieter point: its search and fetch API costs $0.75 per 1000 requests, against about $4…
- DeepSeek Flash undercuts closed labs on price and search bullish
- Local LLMs: people want to try before they buy hardware mixed
- Coding agents are leaving a mess on developers' machines bearish
- AI slop fatigue and the push to verify and understand code mixed
- Ownership vs licensing: bricked products and rented access mixed
- Infrastructure chokepoints keep surprising people bearish
- Hand-built tools and simulators: craft as a counterweight to AI bullish
Saturday, October 17
230 threadsClaude redesigned eight quantum circuits and cut operations by up to 17x, and in a separate thread it traced the origins of an ECDSA curve that had fed 13 years of NIST conspiracy…
- Open-weight models keep closing the gap, and people want them local bullish
- Claude is being credited with real research results, with side-eye mixed
- Benchmarks do not measure the job, and everyone knows it mixed
- AI-generated volume is straining review, code and content alike bearish
- Central chokepoints keep breaking things for builders bearish
- Builders are drifting back to fundamentals and simple tools bullish
- Tesla Roadster demo wows, then the shipping date cools the room mixed
Friday, October 16
230 threadsMeta shipped Muse Spark Open, an open-weight model built to run on Unsloth, MLX and just about every other runner, and HN cheered. Meanwhile, Anthropic's cutoff of third-party har…
- Open weights and local models start to look usable bullish
- Platforms keep changing the deal after you've built on them bearish
- Agents are now doing real hardware and low-level work bullish
- Backlash against AI-generated text is hardening into policy mixed
- Cloudflare and a few giants are the web's single point of failure bearish
- Meta's terms change and the fight over scraping bearish
- Compensation data and tax rules keep tilting against workers and builders mixed
Thursday, October 15
230 threadsA commenter who once audited an Amazon FBA seller's books was stunned by how many fees the platform charges. Elsewhere, a 24B-parameter model hit 184 tokens a second on a laptop. …
- Local models and open harnesses are the hedge against AI vendors mixed
- Platforms squeeze sellers, and builders keep getting burned by dependencies bearish
- LLMs are becoming the interface for physical design and small web tools bullish
- Distrust of doctors and dentists drives DIY research and wishful thinking mixed
- Loneliness is shaping small products and the comments around them mixed
- Low-level craft is still a big draw: exploits, allocators, and 'standard' C bullish
- Niche data-aggregation Show HNs: one painful lookup, turned into a product bullish
- Authenticity under pressure: AI slop flags, account purges, and veterans using AI mixed
Wednesday, October 14
230 threadsGoogle shipped Gemini 4 Argon and a developer in the EU found it was "not available in our org," while Anthropic, in a separate thread, was explaining how it redesigned source con…
- AI vendors decide who gets the model and how mixed
- Anthropic rethinks source control for AI-written code mixed
- Agents break things and slop floods the commons bearish
- Platform dependence keeps costing builders bearish
- Ledger's chip implant ends the "theoretical attack" excuse bearish
- Postgres vector search heats up as memory gets expensive bullish
- Craft and recreational programming push back on AI-everything bullish
Tuesday, October 13
230 threadsCharacter.ai filed for Chapter 11, two years after Google paid roughly $2.7 billion to hire its founders and about 30 researchers, and its CEO is heading to Disney as CTO. The sam…
- AI math breakthroughs get audited and the cracks show mixed
- Character.ai bankruptcy and the acquihire hangover bearish
- Running big models locally is getting real bullish
- AI in professional work: human sign-off as the moat mixed
- Cyber-physical and supply chain attacks keep getting stranger bearish
- Platform risk: billing glitches, blocks and policy changes bearish
- Builders reject slop and reach for craft mixed
- Mozilla bets on enterprise, and the open web worries persist neutral
Monday, October 12
230 threadsZod 4.6.6, one of the most widely installed validation libraries in JavaScript, shipped with a malicious remote-code loader hidden in a commit titled "Fix bugs and improve perform…
- Maintainers and readers push back on AI-made work bearish
- Anthropic IPO skepticism meets the token cost problem bearish
- Terence Tao on Math 2.0: who must understand the answer? mixed
- Who controls your identity: passkeys and fingerprinting mixed
- Supply chain and operational attacks keep landing bearish
- Local compute wins: fast disks, local indexes, local models bullish
- Demand for open hardware is loud, willingness to pay is not mixed
- Craft and heritage get a nostalgic spotlight bullish
Sunday, October 11
230 threadsA Ukrainian strike on a third Yandex Cloud data center showed up as a sudden drop in Cloudflare's bot traffic graph, and the HN crowd immediately turned it into a joke about "AI a…
- Agents need boundaries, and users want to own them mixed
- War is now visible in cloud and bot traffic bearish
- Platform chokepoints keep breaking things, and people flee to open protocols bearish
- AI is eating cheap hardware, and old gear gets a second life mixed
- A human premium emerges as AI slop floods everything mixed
- ML engineers push back on LLM-for-everything mixed
- Prototyping toys and playful builds keep winning hearts bullish
Saturday, October 10
230 threadsCloudflare bought Deno, and the same day a developer posted Talorys, a personal AI agent that runs on Cloudflare's free tier. Elsewhere, Typesafe AI (yes, the Scala company) raise…
- Cloudflare keeps absorbing the developer stack mixed
- Mega-rounds keep landing, and the skepticism is loud mixed
- Coding agents are now the default way to start any project mixed
- Who is accountable when AI does the work? bearish
- Vendor terms and billing surprises are the new platform risk bearish
- Security basics keep failing in embarrassing ways bearish
- Systems tooling is being rewired, quietly neutral
- Compensation data and labor rules: founders keep asking who benefits mixed
Friday, October 9
230 threadsOpenAI had a bad day on two fronts at once. It withdrew three of its mathematical results after critics said the write-ups read like slop, and it fired three safety researchers, w…
- OpenAI's credibility takes two hits in one day bearish
- Cheap, small and local models keep eating the premium tier bullish
- Agents need sandboxes, and the human-in-the-loop part is awkward mixed
- Do you still need to learn to code? The answer is 'yes, and' mixed
- US policy squeezes talent and data just as AI needs both bearish
- Slop anxiety: people want proof a human did the work bearish
- Small, owned infrastructure keeps winning mindshare bullish
Thursday, October 8
230 threadsOpenAI withdrew three math papers on the same day the American Mathematical Society's AHM issued a statement about its October 6 release, and the defense that the proofs were 'Lea…
- OpenAI's math claims fall apart under human scrutiny bearish
- New model releases arrive weekly, and price per task is the real fight mixed
- LLMs are porting whole codebases, and the bill is real mixed
- Fast to build, slow to trust: the durability backlash mixed
- Agent harnesses multiply while subscriptions and local models squeeze mixed
- Open source and communities build defenses against AI slop bearish
- Concentrated infrastructure keeps surprising its customers bearish
- Remembering the human makers while machines take the keyboard neutral
Wednesday, October 7
230 threadsMistral dropped Large 4, a 1-trillion-parameter open-weight model that claims to match GLM 5.3 on DeepSWE. Google shipped EmbeddingGemma 2 under Apache 2.0, and a 2B decision mode…
- Open weights and tiny models keep closing the gap bullish
- Labs squeeze third parties while absorbing their features bearish
- AI-built everything, and a growing allergy to the slop mixed
- Protecting software you wrote when copying is nearly free bearish
- Internet plumbing keeps failing in surprising ways bearish
- AI for science and math gets a trust problem mixed
- Developers are tired, but the solo builders are fired up mixed
- Crowdfunded hardware lies and the cost of broken trust bearish
Tuesday, October 6
230 threadsReflection released Beam, a 501 billion parameter open-weight model, and the first thing the top commenters asked was where the training data came from. That question runs through…
- Open-weight models keep climbing, and the questions get harder bullish
- Cloudflare is becoming the default middleman, and people are noticing mixed
- Platforms are fencing in AI agents bearish
- AI science claims draw excitement and eye-rolls in equal measure mixed
- AI is eating developer tooling, and the incumbents are bleeding mixed
- Old security flaws plus cheap AI equals new exposure bearish
- Geothermal beats the nuclear hype on speed bullish
- Humans want proof of human, and interviews are tightening mixed
Monday, October 5
230 threadsSomeone got a 125B-parameter Qwen model running at 124 tokens per second on a single RTX 4090, and the same day Apple tightened the screws on agent access in macOS 27. Those two s…
- Local models are closing in on cloud coding bullish
- Who controls the device you already bought? bearish
- Centralized systems keep failing in public bearish
- Cloudflare keeps moving into the middle of everything mixed
- AI slop anxiety, and where AI actually helps mixed
- Computing history is being preserved and rebuilt neutral
- Veteran builders find a second wind with AI coding bullish
- Robotics money flows, with a side of cynicism mixed
Sunday, October 4
230 threadsA hobby AWS account showed a $286 million bill, another showed $284 billion, and the same day HN was arguing that every cloud service should ship with a hard spending cap. That's …
- Hard budget caps are now a baseline demand bearish
- Platform dependence keeps biting builders bearish
- AI lab credibility: culture, safety and 'zero concerns' mixed
- Agent memory is being rebuilt as plain documentation bullish
- Authenticity under pressure: AI slop and weak provenance bearish
- AI is making older builders ambitious and blue-collar work look safer mixed
- Craft and efficiency: squeezing more from existing hardware and standards bullish
- Distributed resilience is the new design principle bullish
Saturday, October 3
230 threadsSalvatore Sanfilippo, the creator of Redis, shipped ds4, an inference engine for running DeepSeek locally. Meanwhile OpenAI is building websites inside ChatGPT, and an Ask HN thre…
- Local and open models are becoming a real daily option bullish
- AI labs keep moving into the layer their customers build on bearish
- Agent token bills are a new kind of surprise cost mixed
- Agents want deep system access and the security bill is arriving mixed
- Cloudflare and the pipes are where internet control fights happen bearish
- AI slop is pushing communities toward provably human spaces mixed
- AI coding wins over some veterans while others find it hollow bullish
Friday, October 2
230 threadsarXiv received 40,363 submissions in September, up from 9,869 in 2016, and it just started rate limiting authors. A Harvard physicist posted 36 papers written with Claude, and a S…
- The agent harness is the new battleground mixed
- Local models are close, but not close enough mixed
- AI output is overwhelming the people who check it mixed
- Cloudflare ships everything, and people are nervous mixed
- Vector databases are folding into regular databases neutral
- Security debt is piling up in the boring layers bearish
- Europe wants out of US payments, and dependency worries spread mixed
- Hiring is AI-flavored, and tooling picks quiet winners bullish
Thursday, October 1
230 threadsGoogle's Gemini 4 Argon landed with a $2 per million input tokens, $10 per million output tokens price tag and a 95% discount on cached input, and the top comment was a shrug: it …
- Model pricing is the battleground, and local models are creeping up mixed
- AI anxiety and AI joy, sometimes from the same person mixed
- AI slop is poisoning trust in text, READMEs and profiles bearish
- Chip design gets an AI bundle while hardware people ask for open tools mixed
- Platforms keep breaking promises and builders keep noticing bearish
- Engineers argue about what actually proves a system works neutral
- Dense, tactile, honest interfaces are having a nostalgia moment bullish