464 stories published this day, by desk. Times are Pacific.
Security48 stories
- A staging password went into a Google Doc, and Google's autocomplete found it first
- The customer is genuine and the payment is authorized: 55% of banks say scams dominate fraud
- Berlin's containment choice: two ministries pulled off the state network, no return date
- OpenAI's Computer History writes a plaintext log of the workday. Decide before staff opt in.
- A year of pattern testing says the weak link in ALPR is the model, not the camera
- Apple patches 27 bugs, and another ImageIO code execution flaw is the one that matters
- Oracle's monthly patch drop hit 925 CVEs in August, four times June's volume
- Influence Operations Now Target Construction Schedules, Not Just Elections
- Contained at the edge, offline at the worst week: UTSA's shutdown before 40,000 students returned
- "Work PC" beats DESKTOP-XXXXXXXX: Entra device-join detection needs a new anchor
- The AI security line item to fund first is log coverage, not another agent
- NPS Bought Flock. Internal Paper Says License Plate Reader; Public Statement Says Traffic Counts
- Siemens patches a CAE overflow that lands in the sectors that patch workstations last
- Akira advisory update: $244m taken, one SonicWall CVE, three controls to audit now
- Unit 42's Credential Brief: Hunt The Login That Succeeds Right After The Failures
- Mabna Institute case re-upped with 17 defendants, this time during a war with Iran
- Medusa's patch window is negative: 500 victims, and exploits used before disclosure
- The extortion layer your IR playbook has no page for: 'Ransom Busters' emails victims mid-incident
- MacSync Stealer: the domain list was the output, not the hunt
- Exploited within hours: MLflow SSRF and FUXA auth bypass join the emergency patch list
- Mythos's method, not its zero-day count, is what breaks CVE-keyed vuln management
- Attacker self-infection exposes a 2,000-site WordPress delivery network
- Prevention back to 69%, and a 91-point gap inside one technique
- Eight agents, four days, 1,395 files: the AI intrusion campaign that mostly ran itself
- Copilot told Varonis how to break it, and that is the third one-click leak this year
- Nitrogen's ESXi encryptor is broken, which means the ransom buys nothing
- North Korea's hiring funnel: 60 applications a day, 22 personas, ten jobs landed
- TWINLOOT runs its entire C2 inside Microsoft 365, and your egress filter will not see it
- Rapid7 counted 8,539 high-severity CVEs and 40 exploited ones. Patch coverage is now a vanity metric
- A paragraph beat the agent "mind virus": reading the Anthropic-EPFL preprint as a defensive win
- StubMaker: 16 typosquatted RubyGems packages, and an audit list for Windows dev machines
- Fortinet Buys Virtue AI, and AI Red-Teaming Becomes a Suite Feature
- Google's reference agent approved a $10,000 refund on a $149 order, on purpose
- Ransomware's price point is $10m to $1bn in revenue, and it is not moving
- One VPS, Two SaaS Platforms: A Portal Scraper That Has Not Moved Since March 2025
- Agent memory leaks: benchmark finds up to 69% of user attributes disclosed in the wrong context
- Storm-0501's first move is deleting your resource locks, not encrypting your disks
- The prompt is the evidence: a 3M expert's ChatGPT log became 350 pages of discovery
- Heights Finance says its loan systems held. 1.2 million records went anyway.
- OWASP keeps prompt injection at number one and starts managing the blast radius
- C2Looper puts its C2 inside GitHub, and domain-reputation stacks will not care
- WMIC is gone from Windows 11 24H2 and 25H2, and your wmic.exe rules go with it
- Prompts, shell history, a jailbreak: AI assistants turn up inside a crypto-fraud build pipeline
- Cavern's DNS Coin-Flip: When Google Apps Script Becomes Rotatable C2 Plumbing
- CISA puts Ray on the KEV list, and the exploit path runs through your developers' browsers
- The credential store nobody inventoried: MCP servers now hold the keys to everything they touch
- The model is now choosing the extortion targets, not just writing the malware
- One vendor, 19 million patients: the MyDr breach is a lesson in whose perimeter matters
Build135 stories
- Anthropic streams tool arguments as JSON fragments, so pick a coping strategy on purpose
- Exit 0 is not a health check: three weeks of macOS backups that copied nothing
- A self-healing Odoo layer scored 83.3 percent, and the 16.7 percent is the useful part
- The conductor is the bottleneck: local agent stacks fail at orchestration, not at the workers
- Scheduling a prompt is not monitoring: keep the LLM upstream of the cron
- Palomar registers Lean proofs against a commit, and refuses to referee them
- Your model cannot tell your instructions from the customer's, and that is the whole bug
- Bedrock-RL's real output is a fixed verifier, not a better Minecraft agent
- Notion's agent stack is live, not slideware, and it only changes one of your decisions
- TestMu AI's Agent Assurance grades agents on the files they changed, not the answers they gave
- 45 or 793 tok/s: the same model, and only one of those numbers sizes your box
- A foreign-key cycle where every column is NOT NULL cannot take its first row
- H3's reference path is a different checkpoint, capped at 12 files, and stops at 768p
- The 2-token prompt that billed 39,966 tokens: what a CLI subprocess hides
- Forminator trusts a forged upload: a dropdown flaw exposes 600,000 WordPress sites to RCE
- Every test passed on a QUIC transport that no code path had ever constructed
- GitLab bundles a zero-click GraphQL flaw with a CSRF bug, and only one needs a victim
- Clop's Windchill implant borrows the app's own keystore, so app logs are the only witness
- A $5-a-month monitoring SaaS on Workers, Turso and R2 is a cost datapoint, not a blueprint
- Temporal talks $12B six months after $5B: durable execution is now agent infrastructure
- AI-written code fails the same four ways, and every gate you own reports green
- Your ORM picked the wrong timestamp type, and a migration linter cannot find it
- A 200 OK that means zero: Workday's job API empties out above limit 20
- "Follow the same standard" is how mobile accessibility debt gets onto the books
- Your Nodes Are Not Idle, They Are Reserved: The Requests Trap In Karpenter Consolidation
- Ornith-1.0's benchmarks are fine. Ollama can't parse its tool calls.
- The First Firewall Rule Is a Cutover: One Allowlist Entry, One Dead Production App
- Exit 0 is not a health check: a dead PID held a lock for two hours and nothing complained
- Careful review passed, tsc passed, and the default dispatcher threw on every call
- Next.js 16.3's memory claim didn't reproduce; its TypeScript handoff cut a build by two thirds
- An agent shipped the HP printer driver HP would not: four hours, and a root daemon to maintain
- That 34.3% TREM2 Hit Rate Belongs To Six Agents, Not To Claude
- Dropbox says the limit on AI capacity is watts and floor space, not purchase orders
- Anthropic's protein run is checkable, which is rarer than the 26.8% hit rate
- SemiAnalysis to software teams: your token cost starts at the fab, not the price list
- Multi-agent security wants to be its own field, and your review process assumes one agent
- Amazon's $19.81bn ad quarter: Godin's profit math is wrong, his tollbooth is real
- SciCon subtracts the text-only score at decode time to stop figure QA from guessing
- ChatGPT tops Google's paid-click share at 4.75%, and growth teams should reprice the auction
- Mercari's TiDB gains came from storage knobs, while plan caching went back off
- Stripe paid $7B for a router. Glean is selling the choice not to call a model at all
- HermitMQ's million-message claim rests on a 29-byte header, not the network
- A blind model scores on your vision benchmark, which means the benchmark grades priors
- Epic confirms a native Linux launcher, but the anti-cheat job posting is the real tell
- GPT-5.6 ships as three models, and that makes model choice a deployment decision
- 117 identical errors, zero bugs: when the defect lives in the orchestration
- Your REPL Is Not A Container: Put Free-Variable Checks In CI Before Generated Code Ships
- Your agent's retry logic is reading a timeout as a fact it does not have
- Two mechanisms, one vCPU floor: why db.t3.micro cannot meet a 1-second RPO on RDS
- AWS gives software supply chain its own Security Hub category, with two vendors in it
- The defect tax on in-editor models is a review capacity problem, not a tooling one
- A 14,000-star watermark remover, and no detector to test it against
- Basic Auth becomes a gateway problem: AgentCore's Lambda interceptor keeps the password away from the model
- Four frontier models in four days, and the cheapest number in your agent plan has an expiry date
- Your GPU reports 24GB. Only 7.9GB of it loads a model, and half of that is already gone
- AWS moves agent payments to GA: the plumbing is done, the sign-off is not
- Agent memory has a dose-response curve, and the cheapest dose won the biggest gain
- A Stripe SDK Major Bump Turned One Metadata Lookup Into a Silent Non-Delivery
- Iceberg votes to forbid new equality deletes in V4, and Parquet 1.18.0 lands with two corruption bugs
- Jumio's sub-100ms feature store is mostly a consistency build, not a speed build
- cuML 25.06 spreads UMAP's kNN build across GPUs, lifting training off the single-GPU ceiling
- Fifty hops, one budget: why more GPU capacity won't fix agent latency
- Your MCP tool definitions cost 64% more than your token counter says
- Kimi Desktop's second binary: mutable CDN path, no Windows checksum, no signer check
- A letter grade, not a megawatt, is the binding constraint at Port Washington
- The optional EntityManager is the bug: moving the transaction boundary into AsyncLocalStorage
- AgentWorm's lesson: the agent is the malware runtime, not the payload
- Go's csv reader allocates by design: what a 4 KB-buffer replacement actually buys
- Axonius runs one agent per tenant on AgentCore, and tracks model cost the same way
- Anthropic meters AWS Claude usage in one-cent units, and charges 10% more to pin inference to the US
- The agent said the Jira move failed. It had already committed the write.
- Cursor's undocumented 'desktop' command turns local AI agents into a scriptable control channel
- Your vLLM Manifest Would Boot SGLang Too, And That Is the Problem
- Your Multi-Key Failover Is The Most Expensive Line On Your Coding Agent Bill
- Cloudflare's one-click AI block names GPTBot, not the bot that decides if ChatGPT cites you
- The reason your agent gets worse after an hour is that nothing ever leaves the context window
- Route leak prevention moves into the protocol, and two Tier-1s are stripping the signal
- Claude Code now outruns Copilot roughly two to one in JetBrains' survey of 15,000 developers
- Your Coding Agent Reads .env, And .gitignore Was Never The Control
- Nuclear AI program adds an AI security vendor, and the $60M is not the company's
- Checkr adds a second verification network in Truv, and discloses almost nothing about it
- Six token trackers, no referee: LLM spend dashboards disagree by 2x to 8x
- The uptime monitor was the outage: 288 scheduled runs a day became 15
- 1,400 npm maintainer domains, 18 flags, and one word doing too much work
- Microsoft's MAI-Thinking-1 lands in Foundry, and .NET teams get a reasoning model without Python
- OpenAI's president says open weights will accelerate the threat. His own cyber model stays gated.
- Kubernetes MCP servers hide the delete tool; hiding is not removing
- At 70,000 images, next/image is not your optimizer, it is your outage
- The npm audit that works because it never installs the package
- An empty path parameter isn't a 404. It's a request for the entire collection.
- When a slow link reads as a dead patient monitor, TCP is the wrong contract
- Sixteen wrong comments, one shape: rot is a category, not a fog
- 802.1X proves the client. It proves the network only if someone ticked the box.
- On GKE Autopilot, a NetworkPolicy blocks the API server and Postgres takes the blame
- Do not let the model that wrote the diff approve it: the case for a cross-vendor review gate
- China's accelerator swap makes Cambricon supply, not export policy, your ship-date risk
- Your $1,600 robot dog has a DEVCOM lineage, and that is now a procurement question
- A NIST AI RMF-mapped RAG system for $25 a month plus a third of a cent per query
- The Tokenizer Is Your Real Price List, Not the Per-Million Rate Card
- Gartner: agent inference cost rises 5x by 2028, so budget per workflow, not per model
- The repair deadline that matters is 31 July 2026, not July 2024
- An agent built and deleted a prod stack. The alert fired on time and changed nothing
- SpaceXAI's new Grok terms make "indirect competitor" a deployment risk
- Your inference bill is an architecture defect: declare the task before you call the model
- Repetitive LLM output is three separate defects, and most teams fix one and stop
- Five pods green, GPU at 99 percent, queue up 70x: the Kubernetes dashboard is the wrong instrument
- TypeScript 6.0's strict default would turn your callbacks into a variance audit
- One non-ASCII letter broke four toolchains: path checks belong in CI, not folklore
- 8,664 indexed pages, nine clicks: programmatic SEO now fails at ranking, not indexing
- One component, three honest byte counts: why published bundle sizes answer nobody's question
- Your Magento admin is slow because order state lives in fifteen tables, not because Varnish is off
- Deferred tool schemas cut cost 21% on average, and made one task type 12.3% dearer
- Dual 3090s, no NVLink: the serving stack broke long before the model did
- A one-ULP divergence, caught because a fifteen-year-old test suite refused an epsilon
- Foundry IQ knowledge bases ship as MCP servers, and four behaviours break naive clients
- Whisper's 300ms floor is architecture, not a bug: when live voice needs streaming ASR
- DynamoDB Now Does Vector Search, So Your Sync Pipeline Needs a Business Case
- PrimeNG archived its repo, and Angular teams now own a procurement decision
- A 12MB Go binary bets agent cost control is cache stickiness, not a dashboard
- Before you spend quota on an agent skill, make it pass an eval harness
- Google Play's 12-tester gate is a status check, not a stopwatch
- One alert, two causes, four green dashboards: the day the stack agreed and was wrong
- The model was never the bottleneck: inside a Kannada RAG rebuild that blames retrieval
- Test count told you nothing about the merge gate; 17 attempts to beat it told you something
- The AI tell is installed during fine-tuning. Humanizer tools edit the output instead.
- Anti-bot systems now score the session, which means your proxy pool is not a mitigation
- Your "Index Only Scan" Did 2,847 Heap Fetches: Covering Indexes Are a Vacuum Problem
- Best agent in Databricks' live document-reasoning contest scored 63.3%
- The matte refiner gets a strip, not the car: a runtime limit on learned correction
- Count invalid JSON as a failed classification, and model choice becomes a reliability problem
- Exit code zero, output zero bytes: why a dated marker file is the only honest success signal
- A supervisor rebuilt from spawn_link and receive shows where restart bugs actually live
- A 32 GB collection with a four-hour restore is a scheduling problem, not a scaling one
- Your AI feature fails with a 200 OK, and your dashboard calls that healthy
- Let the model write the CyberChef recipe, let deterministic code do the decode
Invest103 stories
- A 5.32% Long Bond Beats Foreign Buying: KOSPI Loses 7,000 as Nvidia Backstops $105B
- Korea's brokerages added 402 staff in a quarter while mortgages headed to 5.78%
- Korea narrows its AI model race to three, then plans to fund only two
- Korea's COFIX climbs a fourth month, pushing variable mortgage ceilings to 5.89%
- Korea's state forecaster puts 0.6 of a 0.7-point growth upgrade on semiconductors
- Seoul doubles the FX threshold for foreign bank branches to $100 million
- Temporal's reported $12B ask prices the boring part of the agent stack
- Upbit and Bithumb lose half their revenue, and the fee-only model shows its cost base
- Samsung's breakaway union wants 1,000 treasury shares a head, three months after signing
- Korea's leverage curbs moved the chip bet into margin loans rather than out of it
- Austin -26%, San Francisco -6%: Stop Marking Housing at the National Level
- Meta's Oakland trial puts teen-engagement design, not just money, in front of a jury
- Korea drops the 'outside the capital region' bar, and Gyeonggi joins the cluster queue
- Korea's fuel surcharge jumps seven levels to 21, pricing September off August's jet fuel
- The ECB puts EUR 440bn on Europe's AI trade, then drops the bubble question
- FASB settles its 2027 succession early, and the read is continuity not reset
- Who eats the compliance bill: Sacks and Amodei argue thresholds while OpenAI pauses training
- TheBornKorea's 37% two-day run prices 80 Canadian hotels; the MOU covers one Markham site
- A $15M seed bets the AI data center bottleneck is welders, not chips
- The clock, not the ledger: what a 24/7 deposit network costs 30 banks with $10T in assets
- Robinhood's own blockchain is a revenue-mix problem wearing infrastructure clothes
- Citi will custody bitcoin on its equities platform, and plans to hold the coins itself
- Rillet's $1B mark puts a price on ERP displacement risk
- SEC's token on-ramp: a $5M starter round, a $75M tier, and the end of state review
- A 2022 oracle hack starts moving again, three days after Pando shut its books
- Tenth Circuit doubts on Colorado's rate cap leave the national pricing model intact
- Cboe puts a 404% number on the gap between perps and 0DTE options
- Mortgage bankers' fifth profitable quarter was bought with cost cuts, not pricing
- The non-grantor trust moves from the estate department to the income tax department
- Binance sent Russia a client's passport after quitting the market. Exit claims are diligence items
- Compound puts $52M behind an institutional pivot, and $24M of it is not code
- FASB puts stablecoins on the cash-equivalents line, and the question becomes which ones qualify
- The New Jersey check ring filed real paperwork, which makes registry lookups a formality
- 87,000 firms, 875 roll-ups: the succession math in accounting favours the buyer
- Unsolicited HTX dust makes taint an operations problem, not a blocklist problem
- Perplexity gave $200 subscriptions to 360m Indians and now bills $156,000 a month
- The card networks just picked the referee for agent checkout, and it looks like EMVCo
- Your SOX Agent Vendor Has 90 Days To Prove It Is Not A Chatbot
- With CLARITY stalled, plan compliance against SEC and CFTC staff, not statute
- Budget for the tariffs you have: November 2026, not the September summit, is the deadline
- Polymarket's fine print: machines settle 96.27% of markets, judgement settles 0.92%
- Canada restores R&D capital credits as Xanadu tests whether public markets fund pre-revenue quantum
- Four Years Below Housing Bust Lows Makes the Freeze an Underwriting Assumption
- Neuberger puts junk bonds on chain, and 24/7 settlement stops being a free feature
- Tenth Circuit weighs whether a bank's home-state rate ceiling travels with its loans
- Metaplanet buys a Nasdaq listing with 2,100 of its own bitcoin, and writes a template
- Water Rights, Not Megawatts, Are Becoming the Gating Item on New Load
- Your Landed Cost Is Being Litigated By Companies With $306,000 Problems
- Washington drafts an exclusivity clause for its AI supply-chain club
- The biggest cost advisors pay is one they have to reverse-engineer
- Companies are hiring for the close, not the model: audit tops Heidrick's 2026 skills list
- Nvidia buys the landlord: a $1.5bn equity stake in OpenAI's Ohio power developer
- Baidu's AI revenue grew 25%. Its ad business lost more than AI gained.
- Rippling graded 2,100 agent runs per model. The cheap one basically tied the flagship.
- Anthropic puts a number on Claude Code: $13 a developer-day, and a $30 line to argue about
- One $400M round held up crypto's July: 41 rounds and 140 investors, both at lows
- CoreWeave's $104bn backlog is now a margin question, not a demand question
- OpenAI's teen ChatGPT is a compliance product, and rivals will be measured against it
- The Lakers deal now turns on trust law, not basketball
- Congress Loses Its Two Social Security Reformers Six Years Before the Cut
- Ethereum's 66-proposal Hegotá list is a budget, and only one line item is funded
- The 3.1% drug price drop has three engines, and two of them predate TrumpRx
- Importers Get No Vote on a Midnight Tariff: $20bn of Canadian Goods, 50%, Decided Overnight
- Behind-the-meter gas is the data center buildout's real cost: 318 Mt a year
- Seoul blesses Ripple's third bank deal and blocks Polymarket on the same day
- Retail's real signal this week is the gap between big-ticket and off-price
- Klarna's $1bn Quarter: Margin Dollars Up 42%, Revenue Up 27%, And That Gap Is The Story
- The bond market is doing the worrying: 30-year at a 19-year high, VIX at a year low
- Nvidia and Microsoft bet nuclear's bottleneck is paperwork, not capital
- Ondo's tokenized stock book passes $1B, and the collateral is the story
- Kraken's Krak Card is a deposit play wearing a debit card's clothes
- Four years to EUR 400 million: EURC shows demand, not issuance, is the bottleneck
- Physical AI took $47.4B in six months. The risk moved to bills of materials.
- Hormuz is neither open nor closed, and that is the position you have to price
- Tokenized stocks are a market-structure fight, and $33M is the whole evidence base
- The industrial bank door is open again, and the incumbents are writing the entrance exam
- Hong Kong is opening private markets to retail before the plumbing is built
- Korea's Safe Trust does not remove jeonse risk. It moves it onto HUG's books
- France's under-15 ban dies on the arithmetic of age gates: to check minors, you check everyone
- Spinach doubled in a month: Korea's heat shock is a price mechanism, not a weather story
- Groq raises $350M at half its 2025 price, with Nvidia on the cap table
- EliseAI's $3.7B markup prices the boring layer above its entire software market
- Korea's 40% short-term debt ratio is a WGBI artifact; funding concentration is the real tell
- Zhu Rongji's obituary is a bull case on Chinese demand. Test it before you buy it
- The SEC just put a number on the pre-IPO grey market: 46% average markup, $74M, 800 buyers
- 10,037 illegal-lending complaints: Korea's credit stress has moved off the loan book
- LITILIT's 8M euro loan buys a bet that femtosecond lasers can survive a factory floor
- Canada's foodtech seed base holds. Its Series A market, by CFIN's count, is empty.
- Korea cuts one of four sovereign AI teams, and usability did the cutting
- Meta's Design Risk Stops Being A Headline And Starts Being A Line Item
- Four US grievances at once turn a Korea trade spat into alliance risk
- IPO proceeds tripled to $205.1B, but the deal count barely moved
- KB Home is selling choice because it cannot sell price
- DOJ Flew a $165M Ponzi Case Home From Fiji, Which Resets the Affiliate Math
- Toyota's Cellcentric stake makes hydrogen trucking a three-way bloc, and Hyundai's lead a policy question
- Hormuz tolls look permanent: price the chokepoint, not the spike
- AWS Moves Its Asia Chief To Tokyo, And The Target Is The Mainframe
- Europe's sustainability gap is a missing growth cheque, not a missing disclosure rule
- California's billionaires are spending about 1% of their tax exposure to kill Proposition 40
- The commute is the wage cut: 15% of fuel-hit shift workers are turning down work
- Taiwan turns an 11.05% growth forecast into a cash cheque for every citizen
- Wispr's $280M at $2B says the bottleneck is the text box, not the model
- SoftBank's $200M Buys A Retrofit Kit, Not A New Excavator
Product93 stories
- Oakley takes majority control of GraphDB's maker, with an acquisition mandate attached
- Synthefy's $6.5M seed bets numbers need their own foundation models, not a tokenizer
- He scored "tier one" for AI use. His actual pipeline has at least seven jobs in it
- Weave sells itself to Francisco Partners for $650M, about 2.7 times revenue
- OpenAI's $6.7B quarter, $12.3B loss and the case against a single-vendor model bet
- Unitree opens at 1,100 yuan, and the humanoid cost floor moves to Hangzhou
- Plasma boring's honest state of the art is three metres of granite
- Auradine becomes Velaura: $110M at over $1bn for selling watts, not throughput
- Northrop clears E-2D Block II design review, and what the Navy is buying is upgradeability
- EliseAI's mark moves from $2.2bn to $3.7bn on tenant texts and tour bookings
- AUKUS gets a delivery office: Amentum and Laing O'Rourke to sit inside Australia's Osborne build
- A school agenda shipped with "Vitoiis" and a planet named Marc, and no one read it first
- Proton's Yen stops refusing AI and ships Lumo, moving the privacy fight to terms
- Hypercubic's $5.3M bet: the hard part of COBOL migration is reading the code, not writing it
- Apple splits the iPhone 18 launch, and autumn 2026 becomes a Pro-only quarter
- ABC decides a court docket is cheaper than another year of FCC accommodation
- Clean Core's thorium fuel clears 60 GWd/MTU, then buys an audit. The audit is the news.
- Two auctions, one week: the disclosure gap buried in Situational Awareness's $30bn July
- MDA's new interceptor spec is a price: X-Bow gets $10.98M to chase sub-$750,000 a round
- GE Vernova and SHINE argue nuclear accounting belongs in the design, not the retrofit
- Apple's macOS 26.7 beta names a screened 'HomeAccessory' and a wall-mounted twin
- Etched has silicon, a customer rack running, and $1bn raised in 26 days
- A barge, a berth lease and a standards deal: how nuclear gets back into California
- The Valkyrie's real test was the control handoff, not the flight
- X's ranker pays more for replies than likes, and that makes reach a bad proxy for quality
- Ansible Automation Platform 2.7 deletes the RPM installer, and that is the real upgrade cost
- Cloudsmith's cooldown policies make delay a control, and that makes it your decision
- Apple's best Apple Intelligence feature is a text field, not a chatbot
- Baidu's AI line grew 25 percent and still lost the arithmetic
- A 2x LLM bill is not a bug report: token spend is an observability problem
- Beijing, not Redmond, now sets the retirement date for Windows 10's China build
- OpenAI stops a "significant number" of Astra training runs until cyber gates are met
- Fortinet's Virtue AI deal turns agent security into a platform feature
- Omarchy Quattro pre-wires every coding-agent CLI and refuses to name a default
- Apple swaps a per-install EU fee for a 5% cut, and the math on leaving the App Store changes
- The House's bill drafters now spend longer fixing AI drafts than writing from scratch
- The arithmetic of 17,600 attacker actions: human approval was never a plausible control
- DataVita's £300m is underwritten twice: by CoreWeave's 15-year lease and by the state
- White House lets vetted firms hack back and leaves liability blank for 60 days
- Sixty days to un-design a carrier: the order to put steam catapults back on the Doris Miller
- Five meters, one minute: why voice agent budgets should be priced per outcome
- Google's Pet Memory names your pets, then calls all three cats Smokey
- Google tells Pixel suppliers to be out of China by 2027, and Vietnam is the template
- Starlink Mini works fine at home. The plan attached to it is the expensive one.
- Divine drops invite codes, and the Vine revival becomes an open bet on nostr
- The Air Force Is Buying The Drones It Bans, Because DJI Cannot Be Faked
- Microsoft's cure for the crowded right-click menu makes cleanup your job
- The installed robot base, not the humanoid, will remodel the factory first
- A 2027 Pentagon deadline is pulling magnet-grade iron onshore, one ton a day at a time
- Anthro's Kentucky bet says the solid-state bottleneck is electrolyte supply, not chemistry
- Mozilla's Smart Window bets that opt-in AI is a feature, not a concession
- Two review gates, one nudify app: Meta sold the ads, Apple carried the product
- Comcast turns installed gateways into motion sensors, and the entry-level security market shifts
- Two years of jobs data invert the AI risk rankings: clerks shrink, managers grow
- Perplexity's Airtel giveaway bought 56 million downloads. Now it has to bill them.
- Apple's 29 fresh CVEs are unexploited, which is exactly why the clock started Monday
- The Air Force just published its shopping list of threat signatures it wants to shoot at cheaply
- Celanese and Vigor bet the humanoid bottleneck is plastic, not planning
- An AirTag puts Amazon's name on the book-pulping supply chain
- The allowlist read the command name, not what the command would do
- Teleport puts the developer's Linux desktop inside the audit trail, and takes the SSH keys with it
- Commvault triples Cloud Rewind's Azure reach and reframes recovery as a rebuild job
- Portnox adds Defender to its kill switch, conceding agent credentials outlive the login
- Clerq's ten-minute patentability report targets the junior-associate line item
- Sovereignty audits are moving from the region picker to the plane topology
- Anthropic's $65bn run rate annualises one hot month. Its Q2 pace implies $46bn
- Nvidia's $105bn guarantee, not OpenAI's balance sheet, is what makes Ohio's 8 gigawatts buildable
- Apple tells regulators it may collect nothing on third-party store sales
- Sunrun sells home-battery capacity into Voltus's data center program, skipping the queue
- Canonical: Ubuntu on WSL is growing faster, not overtaking the native desktop
- How a short GNET essay on AI slop became a settled fact about children's brains
- Starlink turns on Vietnam with an ownership waiver, a 600,000-subscriber ceiling and a 2027 rideshare
- Australia bought 500,000 home batteries. Wholesale power prices fell by nearly half.
- The "24 million cars" data centre stat is a unit error, and citing it costs you the argument
- Reported dead in May, live by September: Pawtucket's $128,000 Flock deployment
- Einride orders 500 Tesla Semis, betting on a Nevada line Tesla stopped promising
- DOGE broke the FAA's modernization, Palantir inherits it: $12.5B and nothing for people
- OpenAI now guesses your age, and the guess is the product
- NextEra's $3.3bn Japan-backed tranche buys 10GW of gas, not clean firm, by 2032
- Fairphone lands in the US with a repair price list: $89.99 screens, $40 batteries
- Princeton-led study: AI agents can do the engineering of AI research, not the research
- Half the conversations vanish: an independent check on AI usage reports
- CISA gives federal agencies three days to patch Ray, the framework under your ML pipelines
- Contrail avoidance gets a budget: a £5m trial to see if AI forecasts can move aircraft
- Grounded raises $5M and bets its modular vans on any chassis, electric or not
- China's EV price war has left home: one car exported for every two sold domestically
- Counter-drone sensing arrives as a carrier service, not a base install
- The 1GB flash drive did not die of neglect. It died of bill-of-materials arithmetic
- Tesla's Cybercab hits Austin roads with staff aboard: the metric is days-to-paying-rider
- Cheap bug-hunting arrives: GLM 5.3 puts near-frontier vulnerability discovery on your own hardware
- Eight suppliers, 53 bids, £300,000 each: Britain turned AI tutoring into a procurement line
- Pentagon impatience is the roadmap now: $61m for Smack, $500m for Scale AI
- China's 350-ton electric haul truck gives diesel fleet planners a spec sheet to argue with
Leadership29 stories
- Berkshire's $38 billion Alphabet stake is a governance disclosure, not a tech bet
- Agents Took 150GB From Nine Mexican Agencies. The EU Deadline Just Slipped to 2027
- Human-in-the-loop is being retired, and the assurance burden shifts to machine identity
- You run three technology estates. Governing them as one is how costs escape the budget.
- The SEC's Spring 2026 agenda reads like a planning document. Treat it as one
- Codify it or lose it: plan for volatility in US China policy, not direction
- Expressiveness Is A Cost: The Tradeoff Behind Every "Make It More Flexible" Request
- Kalshi's enforcement chief concedes the equity surveillance playbook does not port over
- The man with 48 baseball teams measures everything except the score
- The FCC Put a Number on "Made in America." Matic's Robot Does Not Hit It Yet.
- Rent the ledger, build the screen: where AI actually moved the buy-versus-build line
- Daycare Does Not Break Children's Brains, And It Does Not Fix Economies Either
- Mamdani's tax office says the data was always public, and sent a memo instead of a witness
- Trump's ballroom case is a standing case, and permitting-exposed firms should read it that way
- The 28,000-row test prompt: credit pricing moved the risk from contract to runtime
- Hourly power futures got an FCM door on August 10, and that is the part that counts
- Norway runs its Ukrainian training camp as a sensor, not a syllabus
- Re-baseline AI procurement on cost per completed task, not dollars per million tokens
- Meta paid parenting influencers while lawmakers debated teen bans. Call it lobbying.
- What an AI filmmaker actually owns is the show bible, not the footage
- VECT 2.0 shreds anything over 128 KB, which makes paying its ransom pointless
- The AI bill nobody reconciles: cost per finished task, not per million tokens
- Anthropic's own telemetry: 93% of permission prompts approved. Budget for blast radius, not reviewers
- The $6m verdict that repriced engagement: 29 states put Meta's feed design on trial
- AI spend nearly tripled in six months. OpenAI's own data says tokens do not predict revenue.
- Dropbox's Nevada exit becomes a fiduciary case, and boards inherit the paperwork
- Cost per successful task, not per token: a 2,400-run benchmark reorders the model shortlist
- Two thousand economists now agree on AI displacement. No firm has changed how it counts.
- You Procured Qwen. Your Edge Boxes Are Running Somebody Else's File.
Science56 stories
- Solar got cheap in factories, not in subsidy bills, and a Hormuz shock tests the difference
- Sargassum forecasts need drifters that stay stuck. Plastic fencing did it
- One 320-metre impact explains Deimos's dust blanket, and predicts what a sampler would grab
- A sensor tuned to its own noise beats the textbook entangled state by 0.698 dB
- Golden Dome's centerpiece is the one part that does not exist yet
- Varroa spreads between apiaries, but Australia still fights it hive by hive
- A horse pain detector sets a testable standard for explainable AI
- Drone delivery stopped being a demo: thousands of tethered pod drops a day in Arkansas and Texas
- Wildfire smoke has clawed back a quarter of 40 years of clean-air gains, and desk jobs are not exempt
- Neanderthals in the Nefud: stone flakes, a drying lake, and no bones at all
- Terrain, not the quake, decides which mountains bleed sediment after a rupture
- A Chinese launch institute puts a nuclear option on the record for the asteroid case DART cannot cover
- Forty-two Jomon genomes put a date on Ice Age isolation and a frequency on cold alleles
- Huashan's red cliff paintings survived 2,000 monsoons because of a lime-and-blood mortar
- A genome-folding machine turns out to pick neuron identities, and that changes the Cornelia de Lange question
- NYC's school match undermatches by list, not by cutoff - and the fix is already in the mail
- Krypton drops tantalum deposition to 200C, and with it a qubit fab excuse
- 21 language models, one habit: tell them your politics and they adopt them
- The safety catch on DNA replication now has a structure, and a mutation that breaks it
- Osterman: 35% of US workers are disposable, and gig apps are barely 1 point of it
- Reefs have a speed limit: 5.3 mm a year, and the Indo-Pacific is on track to exceed it
- NASA puts a number on Mars dust: 0.1 milligram per cubic meter, averaged over 24 hours
- AI weather models cannot forecast what they never saw. A hybrid method aims at that gap.
- Narwhal tusks hide two spirals twisting against each other, and the mismatch is the point
- LRO photographs the Falcon 9 crater, and a piece of debris becomes a calibration target
- A rare neuron-like cell may spare dogs a jaw resection
- An FDA panel wants unapproved peptides sold by prescription. That is deregulation, not wellness.
- Machine-found strategies stick with humans only under three conditions, Max Planck experiment finds
- Marsh plants beat salinity at predicting methane, and blue-carbon ledgers were only half audited
- A four-year visa clock for postdocs now rests on one Massachusetts docket
- Quantinuum's 98-qubit Helios lands on schedule, and the question moves to shuttling
- A pyridine reaction that moves the nitrogen instead of the substituents
- The holotype ledger: 95% of new mammals come from the South, 60% of the specimens do not stay
- Yellowstone's geysers run on water, and 15,000 years of lake mud says drought can throttle them
- Kew's plant report ran a flowering detector over 8 million herbarium sheets in a week
- Two zinc MOFs and a process claim: UChicago ties prediction to synthesis in one loop
- AI-designed phages work. The rulebooks that should cover them were written for a different world.
- NASA's next Mars helicopters go looking for ice, and the antenna has to fold on landing
- AI-drawn photonic parts land 500x smaller, and that is a die-area story
- Sex becomes a setting in cloning: a CRISPR cut makes female clones of a male mouse
- Victoria will vaccinate 5,000 wild penguins against H5N1. The hard part is catching them twice
- For the AMOC, how fast we warm may matter more than how hot it gets
- 171 million foodborne parasite illnesses in 2021 put Cyclospora in proportion
- Two robots in orbit will decide whether satellites stop being disposable
- Heritage's Title IX report cites their studies. The scientists say it misread them.
- Nature paper: diffusion outputs often trace back to no single image, and no single creator
- US surrogacy will be banned over the gaps, not the practice
- A 120-million-year reference section gets a rebuild: the Karoo synthesis
- Before he could publish in Telugu, he had to coin the words
- Stewardship is aimed at the wrong half: rich countries use the resistance drugs, LMICs need them
- Three federal health actions in one week, and the compliance clock starts before the evidence does
- Indonesia scales cloud seeding against a fire season it cannot yet count
- Maternal age effects reversed in one rotifer generation, pointing away from accumulated damage
- A handful of Texan pumas saved the Florida panther. Genetics is still off the plan.
- A magnetar may have done what no lab can: bend light in apparently empty space
- Four sessions of memory work cut fear of failure in 180 young adults, and it held six months