AI·
Humans caught 13.6% of dangerous commands. Claude Code's classifier caught 89%.
Anthropic makes auto mode the default in Claude Code on August 14. Its own study says the permission prompt was catching almost nothing.
devtake .dev
Signal for devs. Tech news, leaks, and trends.
Just posted
Related stories
Apple·
Apple is testing Chinese memory chips for iPhones and Macs despite a bipartisan Senate warning
The Wall Street Journal says Apple is qualifying CXMT DRAM for iPhones and MacBooks. Export rules bar the spec work Apple normally does with a supplier.
Hardware·
One model per chip: AMD bought Taalas, whose test part hit 17,000 tokens a second
Taalas casts a model's weights into transistors. Its 6nm test chip served Llama 3.1 8B at roughly 17,000 tokens per second, and AMD just bought the company.
Security·
Metabase's password-reset endpoint handed out admin, and Framework had to email every customer
A CVSS 10.0 SQL injection in Metabase was exploited from August 3. Framework told every customer their data was taken. Tally lost emails and password hashes.
AI·
Uber spent its entire 2026 AI budget in four months, and the rest of the industry is now metering tokens
Uber's year of AI budget lasted until April. A 15-company survey shows per-developer spend running from $200 to $3,000 a month, and finance has noticed.
AI·
Google moved Demis Hassabis out of the DeepMind CEO seat. Koray Kavukcuoglu now runs Gemini.
Sundar Pichai made Hassabis chair of Google DeepMind and Alphabet chief scientist. Koray Kavukcuoglu takes the Gemini org, and Jeff Dean is leaving.
Agentic coding
Claude Code, Codex, background agents, Routines, and the fast-moving race to let AI run developer workflows end-to-end.
Developer tools
Language releases, IDE fights, design-tool shakeups, and the shipping calendars of the tools developers actually use every day. Not just the AI agents on top — the platforms underneath.
AI models
Frontier model launches, independent benchmarks, leaked capabilities, and the open-vs-closed split. Opus, GPT, Gemini, Qwen — every release, dissected.
Show 6 more topics Show fewer topics
Open source ecosystem
Licensing fights, foundation funding crises, the RubyGems fork, Ubuntu LTS cycles, Firefox's AI-assisted security push, and the governance questions the ecosystem keeps punting on.
AI infrastructure
Custom silicon, gigawatt deals, data-center disclosure rules, and the physical capex behind the AI boom. Who's building, who's buying, who's regulating.
Silicon & memory
Foundry politics, memory-industry wage wars, the DRAM shortage, HBM supply, and the physical chips that increasingly decide what ships and at what price.
Security & supply chain
Package worms, compromised CI pipelines, npm and PyPI attacks, hardware-side exploits, and the post-quantum migration. The security beat for people who build with other people's code.
Apple's AI pivot
Tim Cook's handoff, the Siri bootcamp, the Gemini deal, the iPhone 18 Pro's under-display Face ID, and whether Apple can close its AI gap before WWDC 2026.
Regulation & policy
The EU Battery Regulation, Android's sideloading squeeze, data-center siting fights, and the licensing rulings reshaping open source. Where the lawyers set the calendar, not the CEOs.
Trending
AI·
An AI agent built a working RISC-V CPU from a 219-word prompt in 12 hours. Here's what it actually did.
Verkor's Design Conductor agent went from a 219-word spec to a tape-out-ready RISC-V core called VerCore in 12 hours. The catch: it's still a Celeron.
Open Source·
Rust's new rules require disclosing LLM use. A Linux maintainer gives AI patches three seconds.
Five Rust teams ratified an LLM policy that bans AI-written docs and mandates disclosure. Linux leaves it to each maintainer. Both are rationing review time.
AI·
Alibaba's Qwen3.8-Max beats Fable 5 on Terminal-Bench, and the weights go public next week
Qwen3.8-Max is a 2.4-trillion-parameter MoE that tops Claude Fable 5 on Terminal-Bench 2.1 and trails it badly on SWE-bench Pro. It's the first open Max-tier Qwen.
AI·
DeepSeek's new 304B agentic model now runs on a single 128GB workstation
Salvatore Sanfilippo repacked DeepSeek V4 Flash into a lossless MXFP4 GGUF that streams from SSD at over 20 tokens a second. The hardware bill, and where hosted still wins.
Security·
Malicious npm and PyPI packages turn dev servers into Chinese LLM proxies
Aikido found a stage-2 Go binary inside two health-check-themed packages that runs an OpenAI-compatible router routing Claude, GPT, and Gemini traffic through Chinese aggregators.
AI·
Claude Opus 5 nears Fable 5's frontier intelligence at half the price
Anthropic shipped Claude Opus 5 at the same $5/$25 per million tokens as Opus 4.8. It nears Fable 5's intelligence at half the cost, with new effort and fallback controls.
The open-weight race
Qwen3.8-Max, Kimi K3 and DeepSeek V4 Flash are trading the open-weight lead month by month.
AI·
Kimi K3 trades blows with Anthropic's Fable, and Moonshot is opening the weights
Kimi K3, GLM 5.2 and DeepSeek V4 put open-weight AI next to the frontier this month. What each model is good at, and why the benchmarks mislead.
AI·
Anthropic wants Congress to punish Alibaba over 28.8 million Claude queries
Anthropic says Alibaba ran the largest distillation campaign it has caught, using 25,000 fake accounts to copy Claude. Here is what that claim actually means.
AI·
GLM-5.2 was trained on Huawei chips, not Nvidia. The open weights beat GPT-5.5 on coding.
Zhipu AI's GLM-5.2 is a free-to-download model trained without Nvidia silicon. Here's what the benchmarks claim and why developers should care.
The RAM shortage
AI demand has memory sold out into 2027, and it is landing on phone, laptop and console prices.
Gaming·
€200 more for an Xbox Series X: Microsoft's memory bill lands on European buyers
Microsoft's August 1 increase put the 1TB Series X at €799.99 and £669.99. The cheapest Series S rose 43%, and Microsoft says storage and memory now cost more than 2.5x.
Hardware·
Samsung won more than $200 billion of Broadcom chip work through 2030
Samsung signed an MOU with Broadcom worth more than $200 billion through 2030 across memory, 2nm foundry and advanced packaging. Nvidia locked up SK Hynix the same day.
Hardware·
Google blamed the Pixel 11 price hike on RAM. A gigabyte now costs $12, up from $2.80.
HBM stacks for AI accelerators outbid the DRAM in phones and laptops. Qualcomm, Roku and Google have all repriced, and IDC sees PC prices up 18.3% this year.
iPhone 18 Pro
Apple's September 9 event is close, and the leaks now cover the camera, Face ID and the foldable.
Security·
World Leaks stole 630GB from Tata Electronics and leaked Apple's iPhone 18 Pro files
World Leaks stole roughly 630GB from Apple assembly partner Tata Electronics and dumped iPhone 18 Pro supply-chain files online. India opened a criminal probe.
Apple·
iPhone 18 Pro: a thicker camera, a bigger battery, and a Face ID gamble
The latest iPhone 18 Pro leaks point to a fatter rear camera, a wider A20 Pro memory bus, under-display Face ID, and a battery jump that may be smaller than the hype.
Apple·
Apple's foldable iPhone was stalling on its hinge. New leaked cases say it's back on for September.
Leaked case renders and design mockups put the foldable iPhone Ultra back on a September track, with a near-creaseless screen, dual rear cameras, and a possible white-only finish.
AI·
Claude Opus 5 nears Fable 5's frontier intelligence at half the price
Anthropic shipped Claude Opus 5 at the same $5/$25 per million tokens as Opus 4.8. It nears Fable 5's intelligence at half the cost, with new effort and fallback controls.
Security·
60 hours of AI cryptanalysis. HAWK's authors pulled it from NIST's post-quantum race.
Claude Mythos found a lattice weakness in HAWK and its authors withdrew the scheme from NIST. Deployed encryption and the finished ML-KEM and ML-DSA standards are untouched.
AI·
Anthropic just dropped its Claude Code workshop tapes. The playbook is better than the marketing.
Boris Cherny on Claude Code, Applied AI on prompting, Erik Schluntz on vibe coding in prod. Three Code with Claude tapes hit YouTube ahead of the 2026 conference.
AI·
Matt Lenhard found 49 relays reselling OpenAI and Anthropic tokens. The cheapest runs 97.8% below list.
Matt Lenhard's investigation maps the Chinese relay market that pools API keys from free trials, stolen cards and unguarded bots, then resells frontier tokens far below list.
Open Source·
Warp's terminal is now open source. The cloud agent platform Oz is the actual product.
Warp released its 36k-star Rust client on GitHub under AGPLv3 on April 28. OpenAI is the founding sponsor and Oz keeps the bills paid.
Open Source·
65.9% of the web platform tests now pass in Servo, but the Rust engine can't select text
Servo 0.4.0 landed a record 558 commits and fixed layout on real sites. Its own dashboard reads 65.9%, CSS Grid sits at 37.8%, and text selection isn't shipped yet.
AI·
Ten decade-old math problems fell to an unreleased OpenAI model, for about $2,000 of tokens each
OpenAI published ten results in math and theoretical CS from an internal build of Astra, with Lean 4 certificates for every proof. What that verification does and doesn't settle.
Android·
Android may stop letting a phone debug itself, and Shizuku would break with it
A comment on a Google issue tracker floated binding ADB to the Wi-Fi interface only. That one change would kill loopback debugging, Shizuku, and every debloater built on it.
AI·
Running a coding agent fully on Apple Silicon, no cloud, is now an off-the-shelf stack
A popular Hacker News how-to walked through a fully local coding agent on Apple Silicon. Here's the realistic 2026 stack: runner, model, and harness.
AI·
Chinese proxy networks sell Claude API access at 90% off. They harvest every prompt that passes through.
A ChinaTalk investigation reveals how 'transfer stations' resell Anthropic API access using stolen credentials, model substitution, and prompt harvesting.
Web·
Vercel's AI SDK 7 makes agents a first-class building block for TypeScript developers
AI SDK 7 ships agent classes, provider-agnostic reasoning control, durable workflows, and a Node 22 + ESM-only baseline. Here's what changed and the migration cost.
AI·
Google shipped three Gemini Flash models but held back its flagship Pro
Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and a cyber specialist, and teased Gemini 4. The 3.5 Pro tier it promised in May still isn't out.
AI·
GitHub ran four frontier models through Copilot's harness. None won every task.
GitHub benchmarked Copilot's agent harness against Claude Code and Codex CLI on five tests. The token savings are real, and the best model depends on the task.
Web·
TypeScript compiled to native: Vercel's scriptc ships a 178KB binary with no JS engine
Vercel Labs open-sourced scriptc, a TypeScript-to-native compiler with no JS engine in the binary. Its README claims 2.4ms startup. Hacker News is not convinced.
Hardware·
Nvidia wants to own every chip inside the AI data center
Nvidia already dominates AI GPUs. Now it's pushing to supply the CPU, the network, and the whole rack. Here's who that squeezes and whether buyers benefit.
- Agents that start with access to nothing: Cloudflare open-sourced the agent platform it runs internally
- Anthropic is building an in-house chip team to run Claude cheaper
- Meta's Muse Code: a terminal agent built for repos most devs never touch
- Atlassian went quiet for 75 days on a Rovo bug that leaks Jira tickets