373 stories published this day, by desk. Times are Pacific.
Security2 stories
Build179 stories
- Blocklisting by extension ID has a shelf life: Open VSX cleared three names in five days
- A FinOps MCP server's safety property is the absence of a write handler
- 85% of failures alerted nobody: what an opt-in safety net costs at 225 workflows
- The i18n decision you make before translating: URL shape, and why ?lang= cannot be undone
- Averaging 44 checks is how a noindex ships: the case against audit health scores
- Kubernetes spreads pods where it likes; a few manifest lines and some arithmetic decide the rest
- The 2 AM gap is not missing logs: it is logs that cannot name a trace
- Read-only or nothing: the one line Allspaw and Long draw around AI in incidents
- Two design mindsets, one LLM policy, and the reason the new rules cancel each other
- Tailscale's 19 corrupt SQLite files show single-writer discipline is not corruption detection
- A 200 Is Not A Row Count: How Agents Inherit Open Data's Silent Caps
- Redlock's five masters are the price of a lock that survives failover
- Pick the BAA, not the benchmark: where HIPAA document pipelines actually spend
- The open port and the egress bill are one ticket, not two queues
- Coding agents cost $4,125 a month because 73% of it is context you already sent
- A GitHub graph in 21 requests: what PR metadata already knows about your files
- 366 green tests certified a contract that mandated a contradiction
- Two builds, one instruction stream: where GCC actually records a non-executable stack
- Codex's usage wall is being rebuilt as a cheaper tier, shipped binaries suggest
- Every PNG optimiser filters by default. On seven of eight animations that made the file bigger.
- Not Responding is a message-loop state, not a fault: get the evidence before the reinstall
- The false green: a cached "connected" cleared a live health check before any model ran
- Know when to skip the cheap check: a $0.002 rehearsal can authorise a $15.50 mistake
- ToxicPanda 2.0 talks its way to an ADB shell, so wireless debugging belongs in the baseline
- The bug is the tutorial's first line: "set up your vector database"
- Wan 2.7 puts the audio inside the render, and takes your editing seam with it
- A green backup job proves the write; only a second instance and CHECKDB prove retrieval
- ZeroTier without VL2: an ESP32 can peer through NAT and inherits five reliability jobs
- Four clocks, one number: what a Laravel credits package takes out of usage billing
- Meta's omnilingual ASR: 1,693 languages, 17GB of VRAM, and a capacity problem instead of a vendor list
- sbi swaps three network-spec routes for typed configs, so bad settings fail where you write them
- JDK 27's nine JEPs are frozen, and two of them change your defaults
- The fine-tune that kept promising refunds: freshness is the bug no eval catches
- Six-figure map markers: the primitive is a vertex buffer, not a DOM node
- SSE is 30 lines on one pod. The second pod is where the engineering starts
- A retro that files but cannot gate: three fixes proposed, none built, health check green
- Prompt injection has no parameterized query, so stop shopping for a sterner system prompt
- One state machine or two: the transaction fork hiding inside your MCP server
- Cheap GPU-hours, expensive finished jobs: what the comparison page leaves off
- The importer said it worked, wrote nothing, and left two accounts behind
- Shopify flipped agent access on for you. The only decision left is what you log.
- Xiaomi's 2,699 RMB four-bay NAS prices the chassis at 675 RMB a slot
- The fix for a confused coding agent is a smaller input, not a bigger window
- 56 build-pipeline attacks, one vendor's alert queue, and the February jump nobody can attribute yet
- Replit's August 21 bundle sells scheduled agents; the governance half is still a label
- Neither can build it alone: Korean memory and Taiwanese packaging now set the accelerator clock
- Every test passed: the caption guard was measuring a string the renderer never drew
- 99% token accuracy, zero learning: the training failure that bills you for 18 hours
- Three outages, four alarm slots: what a nightly log drain teaches about laptop scheduling
- 157 agent runs, 18 configurations, and the one variable nobody actually tested
- A recommender whose items get tired: expert matching fused by rank, not by score
- Nous wants you to own the harness, which means you own the patching too
- The hard part of running five coding agents is not the model, it is the process tree
- A 192-row intent dataset with a DOI is worth more than your bigger one without
- wmic is gone, and Node's process trees on Windows 11 now come back empty
- Cloudflare OS makes the sandbox the product, and the document the unit of isolation
- Parallel coding agents on Windows break at the home directory, not the launcher
- Two tokens, one exchange: binding a user's consent to an attested agent workload
- A $10 mesh robot moves the classroom bottleneck from the purchase order to the radio
- On-device continual learning gets an artifact: 230M model, rank-4 LoRA, one frozen teacher
- Four of six steps can bin a resume before a human sees it, and the worst is an image-only PDF
- Code review's quality gate survives AI. Its apprenticeship does not.
- One rule for a day of bug fixing, and a restart planner that argued with its own baseline
- Android has five storage models, and your path strings only work in one of them
- Ninety Milliseconds Of Work Inside A Ten-Second Request
- The gate that fired on its own code: check that you were allowed to generate, not the output
- Webpack's docs now build from its own types, and the bill is 15 PRs across two repos
- The CI gate that exits zero: wiring MCP tool-poisoning checks so they fail the build
- Ten tasks, three runs each: grading a free coding model before it edits your repo
- A cascade router beats a retry loop on 429s, but its breaker never actually trips
- Hugging Face's $13B process puts most teams' model pipeline under a single owner
- Client-side transcode moves the GPU bill. It does not delete it.
- AgentCheck reviews what the agent actually changed, and does it without another model
- Atlassian's full AI opt-out costs an Enterprise upgrade: three of four tiers can't switch metadata off
- Seventeen fields written, zero delivered: why "done" has to come from outside the agent
- Before you restart that Windows service, find out whether StartType survived the update
- Exit code 137 is the kernel collecting on a bet you did not know it placed
- CAPTCHA walls in Playwright agent pipelines are usually self-inflicted retry bursts
- Calibrate once, route forever: the cascade arithmetic says the cheap path carries under half
- PyRIT went read-only in March. The four tools left do not overlap the way the guides assume.
- The re-encode that quadrupled a GIF was passing every test that looked at pixels
- .NET 10 fails restore on a versionless PackageReference, and the reflexive fix breaks CPM
- An LZ4 cache pitched at np.save speed writes at about 1.5x np.save, by its own numbers
- The A2A card said 0.0.0.0:8080, and the only caller it broke was Google's
- The pre-migration subnet plan is an audit against reality, not a diagram
- Fable 5 at $50 per million output tokens turns model routing into a budget line
- Search Console said delete nothing: 428 generated pages indexed, 2,290 handwritten ones never fetched
- Three clouds, one protocol, no stored credentials: the runtime becomes a swappable part
- Express Mode builds the ALB you skipped, and owns the knobs you used to write
- The five-minute grep: what `synchronize: true` tells a reviewer before your tests do
- The 300-monitor wall is real, and sharding Uptime Kuma is a bill you keep paying
- A token budget is not a stop condition: 2,000 lines for a 40-line job
- One join point, two audiences: why the MCP server reads the build artifact, not the source
- Four field types encrypted, one left readable: the decision that prices per-user envelope encryption
- The dofollow lists are right about rel and wrong about everything above it
- The line JavaScript cannot cross, and who pays for going around it
- Adding a reviewer agent bought one useful objection in five, and two new bugs
- The INP bill goes to the wrong department: ad script, not RAM
- A moderation score is evidence; shipping it as a verdict is how a classifier rejects an applicant
- Agent memory under a million rows: a timestamp column beats top-k
- The weights never moved: what 6,852 Claude Code sessions say about where regressions live
- OpenClaw makes the channel the architecture, and the reasoning loop a lodger
- Europe's arms plants are burning, and the arsonists are being hired locally
- TP-Link's $399 Costco mesh is cheap per node, and the rest of the bill is unpriced
- The model was fine: a 3-second P99 at 740K ops/sec was reads queueing behind writes
- A 27B model reportedly beat a license check in 30 minutes. Nobody has seen the binary.
- In Manifest V3 every message handler is a cold start, and the fix for lost state hides new failures
- Agreement to 5.5e-16 across 8,000 boards did not stop a solver calling safe squares mines
- 912MB to 108MB is mostly typing now, and that weakens the base-image excuse in review
- A key in the app binary is a bucket handover; presigned uploads also drop the proxy data bill
- Object or file storage: the wrong pick stays quiet until inode exhaustion
- Velocity was stable because the work was: 31, 34, 32, then 14
- The Slack CLI that skips admin approval keeps live tokens in a file your agent can read
- The weighting ran, then the dedup pass undid it 56 lines later
- Three Russian clusters phish the grant, not the password, and MFA completion changes nothing
- A green suite, a $0.05 ceiling, and the 25 places it was separately enforced
- Harness-Bench makes the scaffold a measured variable, and model-only scores lose meaning
- Go 1.27 gives methods type parameters, and one line about interfaces decides who benefits
- GitLab ships five security fixes and keeps the details sealed until October
- Kubernetes 1.35 lands 60 enhancements, 17 of them stable: treat the upgrade as a compatibility audit
- Excel validation is a gate on one route, and paste walks off with the gate
- webpack stopped hand-writing its API docs, and the hard part was the credential
- Domain screening is not mailbox verification, and the difference decides where policy lives
- Spine Swarm makes the canvas the coordination primitive, and moves the hard part to the validator
- Calibration as position sizing: one over-confident asset, six under-confident, no single fix
- Two regions, 86ms apart, 28 TPS: the WAN was not the bottleneck, Python was
- A five-hour watermark remover, 14,000 stars, and no detector to test either side
- The FCC read HOVERAir's Versa by what it was built to do, not how it was boxed
- RPA is a workaround for a missing API, and buying it otherwise is a category error
- A VPN server that hosted its own iDRAC turned a failed reboot into a 500km problem
- Fixed-window rate limits hand an attacker 200 requests in two seconds
- Spring Boot Actuator: the wildcard is the bug, not the endpoint that sounds scary
- Test the rule change, not the model: a cheap A/B harness for agent standards
- The smallest serious coding agent has a 443-line edit tool, and that is the point
- The AI boss forgot its own handbook, and humans had to hand it back
- ZizkaDB bets agent debugging on edges you declare, not spans you read
- The failure mode is silence: what a weekly per-line output report catches that uptime does not
- A build-tool swap that took 70 days and 166 files, and the build was the easy part
- Dividing by possession made the model worse, and only the second score showed it
- Two phones, no Wi-Fi: the local multiplayer case every tutorial leaves out
- The manager readme failed because the document was never the deliverable
- Two workers, one limit: a rate limiter's storage adapter is its outage policy
- Unitree's 8,000x book is now the comparable every humanoid banker will quote
- Agent memory that learns from wins is grading the user, not the context
- Arid 2.0 keeps the detector and rebuilds the plumbing around it
- The saved-hour fallacy: a model in which better AI makes each paper worse
- Pi ships four tools and no sandbox, so the guardrails are on your build list
- A green suite, a dead server: the MCP bug that lives four lines above your tests
- Partial indexes on Postgres queues: 41x smaller, 6.9% faster, one bound parameter from 7 tps
- The enum that was never written: COMPLETED for runs that did nothing
- Every MCP server you add costs about 11,000 tokens before anyone types a word
- pgrust's 5-microsecond compile drags JIT onto the short-query path
- Death threats are now a siting input, and 500-plus towns are the output
- Iceberg won the table format war, then left the maintenance layer to you
- Strict mode is a migration, not a checkbox: MySQL truncation and the 1406 that follows
- A PDF has no words in it: 'edit one word' is not a string operation
- Ten MCP servers, 847 tool schemas, 112K tokens before you type: curation is a cost line
- OVHcloud reprices most of its catalog as its own RAM index hits 6x, with 9x due in September
- Next.js 16 moves the cache decision into the component tree, and the review surface with it
- The compiler checks three of the seven things your coding agent had to get right
- A reviewer that rejected 86 of 112 repairs still scored 75 percent on its own safety test
- Next.js checks Origin for Server Actions. Move the same logic to a route handler and it stops.
- The soft 404 that returns 200, declares lang=ar, and renders English
- One status code, two opposite remedies: the 429 that cost ARGUS seven minutes
- The missing field is the product: why agent listings need declared, derived and unknown
- Forty animated backgrounds, one battery budget: how Pairly rations frames on old Android
- Two owners, one cancellation source: why async completion should come last
- Same Gauss-Kruger zone, two datums, 427 metres: read the proj4text, not the label
- A buildless app still has a build step: 41 pinned URLs and a sed pass that missed two
- Claude at ten percent of list: China's proxy market makes geoblocking a billing control
- GuardDuty says exfiltration and the patch is four hours out: revoke the sessions first
- Probes cannot fix bad addressing: what a browser-run Kubernetes cluster actually shows
- Docker writes its own hypervisor, and inherits every bug in it
- Firecracker won't run on a Mac, so Encore rebuilt the Linux image toolchain to match
- MCP is four trust boundaries, and credentials only close one of them
- Nine bundled CLIs, zero static links: an OpenSSL CVE becomes a file copy, and the parser is the bill
- Buildroot's two-command ARM image, and the directory that will brick your board
- Angular's compiler already knows the line number. Nothing ships it to the browser
- A Bluesky starter pack is three records, and only the metadata edit is repeatable
Invest130 stories
- Bitcoin's record $14,264 week arrived before the flow data that is supposed to explain it
- Zondacrypto's exit was on-chain for months before the site went dark
- Europe trades cheap because its firms cannot scale, and that does not mean revert
- Phantom prunes Sui and Monad, and the exit fee points to Solana
- Two judges, 42 hours: Nvidia's print and Warsh's first keynote price the same trade
- Pendle rented $50M of lender depth on Morpho, and the price is in the emissions schedule
- Samsung moves heat pump output from China to Gwangju, and the subsidy does the heavy lifting
- A conditional charter for the president's stablecoin, with Circle and Coinbase in the same queue
- Kushner routes Jeffries to the chief of staff, which is what pricing in a lost House looks like
- Beijing's biggest funding push is arriving late into a 9.4% private investment slump
- Anthropic cut 80% of Claude Code's system prompt and the evals did not move
- The $200 AI seat is a subsidy: ceiling users pull 40x to 70x what they pay
- Memory inflation reaches the rack: Nvidia's 17% hike lands on Microsoft, Google and AWS
- Incheon buys 1,300 parking bays for $108m, and 8,503 more at a price it has not published
- Nexen's 4MW power contract shows who pays for the carmakers' Scope 3 targets
- The market bought a superlative for $2.20 a barrel. Bessent's fine print has to beat it
- Socar bets 150 billion won that its 40,000 crash videos a year are what AV models lack
- OpenAI says the attacker is months behind. That makes defense a rental, not a purchase.
- Bitcoin's 23.5% week was a Treasury trade, and Treasury can take it back
- A 5 nm gold sacrificial film turns a 2D materials clean-up problem into a process module
- EMEAP's scam task force moves fraud data sharing from goodwill to supervision
- A $598M short book on one venue, and 24% of it is bearish the venue's own token
- Shein's $100M Everlane deal shows the US gate is now geopolitical, not antitrust
- Korea's weight-drug depression finding rests on 846 users. The 4.2x ideation number on fewer.
- ECX's one-for-one Bitcoin snapshot is a custody deadline before it is an investment question
- IBK's China unit lost 83 billion won because one platform held both the cash and the ledger
- Korean golf's hangover reaches the P&L: PXG operator's operating profit halves
- An mRNA vaccine cleared Phase 3 on Keytruda, and the read-through lands on Alteogen
- TVB's studio campus, Gaw Capital's HK$2bn, and the legacy owner as AI landlord
- FinCEN's CTA rollback hands beneficial-ownership diligence back to lenders
- Ottawa will not talk until November, so price the 50% tariff as a standing cost base
- Seized Wegovy parcels outnumber licensed shipments nine to one in Korea's customs data
- The New York Times starts answering queries the way the answer engines do
- Two interventions, two round trips: the long end now sets the discount rate
- Tungsten up 622%: the AI buildout's chokepoint is an 8,800-ton gas market
- Walmart's Scenario bets under $25 that its apparel problem was style, not price
- Bitcoin's 23% week ran with rising yields, which breaks the risk-on model
- Palantir crossed $1bn in federal obligations. Price it like a prime.
- Hui Ka Yan gets life, Evergrande gets fines worth 0.8% of its $300bn default
- Nearly nine tenths of Anthropic spend is not on its best model, weeks before a $2T listing
- Shein's Hong Kong IPO asks 73% less than 2022, and insiders may buy half the book
- Helcim raises $38M on a bet that Canada's big banks are done with merchant acquiring
- USD1 leaves BitGo for the OCC, and the Senate's CLARITY arithmetic gets harder
- Fifty percent on $20B of Canadian goods: 17 days to re-source lumber, steel and appliances
- Bessent's 'Economic D-Day' puts the Iran sanctions bill on intermediaries, not Tehran
- Iran calls third-country compliance an act of war as Bessent readies new measures
- Lianlian puts agent-wallet controls in a stock filing, on a 24-basis-point take rate
- Binance is still verifying EU users 49 days after MiCA's deadline, tests indicate
- Two labs, one venue: the AI lobbying money tells buyers to plan for state rules
- Canada's $20B counter-tariffs land Sept. 8, and the de-escalation trade is dead
- Iran policy moves to Treasury, and the front line becomes your counterparty file
- Circle's 84% and the card networks' flat year: check the arithmetic before the thesis
- UBS's 8,100 target is a bet on $400 of 2027 earnings, and little else
- USDT wins users 2.7 to 1, USDC wins the venues: distribution is setting stablecoin share
- A 4B model edged the GPT-5 family at bargaining. The interesting number is the spread
- Iran points the Hormuz threat at its neighbours, moving the trigger out of Washington's hands
- Fed disunity, not the Fed's path, is the risk in duration before Warsh speaks
- Korea's crypto market: 566,352 foreign accounts, 90 users, one missing bank account
- Korea's research funder is building the unit that will grade the universities it pays
- Shorting bonds needs three forecasts to land at once, and pays rent while you wait
- Price the cadence, not the level: $1 trillion of new US debt in under five months
- Korea drops Calculus II from its 2028 exam and hands the gap to engineering faculties
- The SEC finally wrote the rule: 401 pages, two dollar caps, and one blank threshold
- 63% of Amazon's religious titles look AI-written, and the only gate is a checkbox
- Central banks concede the CBDC answer, and the stablecoin fight moves to issuers
- Treasury is managing $40 trillion by routing around its own bond market
- Warsh's Silence Has a Price, and the Long End Is Paying It
- YouTube's chart-stream paywall moves retail's free market feed into channel memberships
- Nvidia's implied 6% against an 11% bear case, with Warsh due two days later
- The Fed's grip on the belly of the curve is a maturity ladder, not a rate view
- UniCredit at 48%: control of Commerzbank now turns on the rulebook, not the premium
- Turkey eases three points without calling it a cut
- Hormuz gets a price list: Iran turns $2m shakedowns into a transit fee with an office behind it
- Kashkari's long-end diagnosis: a supply problem the Fed does not plan to solve
- Crypto's marginal buyer is a momentum function, and the Cleveland Fed put a number on it
- A shared vendor is not a Fed endorsement: reading the XRP-FedNow claim properly
- Canada prices the cost of hitting back, and hands cross-border shippers a dated calendar
- Term Labs' $8.5M drain needed no bug, only the votes depositors never claimed
- Goldin's birthrate paper names a variable that no budget line can buy
- Anthropic leads OpenAI by 2.9 points overall, 7.4 among the heaviest AI spenders
- Nine bossware apps, 121 exports of worker names: monitoring is now a vendor-diligence problem
- Everyone has an AI roadmap. Almost nobody has a headcount disclosure.
- The $100bn needs $2tn in state: California's billionaire tax is eroding its own base
- Amazon stopped eating the memory bill, and the Echo Dot went up 60%
- The AI trade's weak link is the buyer: Anthropic's best model took 6% of its tokens
- Douglas reopens to Mexican cattle on Aug 24, at a fifth of pre-ban flow and on revocable terms
- Bitcoin's 20% week was a Treasury trade, and gold is the proof
- Washington's biggest Iran sanctions yet meet Iran's biggest oil customer saying no
- Dollar carry's longest run since 2008 reads as a crowding signal, not a return
- Treasury books a bid at the long end through November, and that is the message
- Nvidia's August 26 print: 92% of the quarter rides on one segment
- Tokenized equities now have a duopoly, and one side holds 85% of the flow
- The AI deal frame flipped: buy at 15 times revenue, pay with paper marked at 40
- Hanwha's $299m US injection is larger than the contract that prompted it
- Microsoft's ThinkingBox puts a number on agent reliability: 65% once, 25% every time
- Seoul's sub-500-unit approvals would move from one mayor to 25 district chiefs
- Advisors are not calling the top. They are pre-funding the withdrawals.
- Seoul's economists want capital allocated for survivability, not export volume
- Camp Kim's housing unlock is a green-space edit, not a Yongsan settlement
- Korea's ruling party pulls the residency test out of its own property tax bill
- Bigger Treasury buybacks, same long end: 30s near 5.25% make duration the holder's problem
- Nvidia into August 26: the beat is priced, the $210 line is not
- Europe's heat toll is now a budget line, and nobody has published the invoice
- Oh Se-hoon names a substitute: Yongsan turns from a veto into a parcel swap
- Two ETH bought a supermajority: Term Finance's $8.5M loss was a vote, not a bug
- The real prize in Korea's 32-trillion-won power merger is one internal division
- Tokyo's 3.8% assumption: debt service eats two-thirds of Japan's budget increase
- Korea's fabless lobby wants orders, not grants: NPU quotas and defence set-asides
- When the marginal bidder is a billionaire, farmland stops being priced off what it grows
- The SEC repriced the raise, not the reporting: 2026's two moves and your listing plan
- Beijing's humanoid games: 2,056 robots, mandatory autonomy, and a 40,000-unit supply base
- Yeouido asks crack: Korea's Aug. 3 tax overhaul is reaching owners, not just sentiment
- A back-to-back Bank of Korea hike is now the base case, not the tail risk
- Korea's 80% youth mortgage lands on a housing stock that barely qualifies
- Seoul court weighs putting creditor talks before the examiner's valuation
- Glass, not lithography: Korea's HBM lead rests on a packaging step it does not own
- Regulators head for Sejong, and the access bill lands on the firms they supervise
- Singapore's deepfaked cabinet cost one victim S$4.9m, and the NDA did the real work
- Hyundai union stands down one walkout, keeps the next: 120 idle line-hours and counting
- The SEC's Walter probe says private credit's risk is who owns both ends of the loan
- Anthropic's discounts leave a 7.5x-to-37.5x gap, and the routing code decides the rest
- Moscow refuses a shipping-only truce, and the Black Sea premium stops being seasonal
- Muan County takes its seat, and an 800 trillion won fab cluster gets a committee date
- Korea's liquor shipments sit 2.3% above the 1998 crisis floor. Zero-alcohol is a holding action
- Korea's grocery money is moving to the corner produce shop, and promotions cannot price it back
- Korean banks' margins widened, profit fell: the swing factor is the bond book
- Over 1,000 trades in June, the big ones on one day: what the Trump filings do and do not show
- Visa's Korea-Japan spend data points at secondary cities. It cannot price them.
- Japan sent 2.53tn yen abroad in one week, and equities took the bigger half
- Iran's negotiators want out; the Guard does not. The oil premium now tracks a factional fight
Product36 stories
- The open-weight default is for sale at $13B, and its buyer pool ships models too
- Mophie's PowerStation Plus bets a $29.95 floor and a fixed cable beat being small
- Spotify and Apple Music will label AI artists within weeks. That is a metadata problem, not a debate
- Spider-Man art book: when clone-stamping reads as AI, a policy is not evidence
- NNSA's first Mk4B arrives three months early, which is a claim about the factory, not the warhead
- Galbot's 100-rally tennis robot is the best whole-body demo yet, and every number in it is Galbot's
- An agent that obeyed its brief and hacked Hugging Face: why permissions are not intent
- GTA VI's $79.99, disc-free preorder sets the ceiling everyone else prices against
- Buc-ee's beaver suit: policing a mark got cheaper in court and dearer everywhere else
- 660 pounds on four legs: DaxAI's robot horse is a spec that needs a slope test
- Nielsen puts Miles Morales ahead of every live-action Spider-Man, by 1.7 billion minutes
- Waymo's 5nm ingest chip moves autonomy economics into the silicon design budget
- Apple is moving store furniture, which says more about the home hub than any parts leak
- Marines buy narrative analysis sole-source, and set the pedigree bar for rivals
- Apple's foldable gets praised for its hinge, and nobody is quoted on the pictures
- The Air Force bought the charger, not the vehicle: Xos's first defense award prices the bottleneck
- Flock wants a privacy-safety compromise. The 46 misuse cases are an access-control bug.
- Alsup fined the download, not the training: the copyright risk in your stack is provenance
- Councils keep a quarter of the ideas only one model had
- Chang'e 7 slips to 2027, and "does not meet the conditions" is doing a lot of work
- Docker Desktop 4.86 swaps libkrun for Docker's own hypervisor, and keeps your old setting
- A 21km air gap in New York's quantum testbed, and the wavelengths it buys
- Asus fixes the one OLED flaw desk buyers cared about, then coats it in a mirror
- LiDAR frees the robot mower from the sky. At $3,200 it is still priced for the biggest lawns
- A $22.3m counter-drone contract that buys warning time, not shootdowns
- A humanoid beat Bolt's 100m mark on a course that removed everything hard about the job
- Rosatom counts 15 small reactors built and 13 in build. Most of them are icebreaker parts.
- A long press in ChatGPT is now the fastest visual assistant on an EU iPhone
- A fair fight with last year's iPhone is where Pixel hardware has landed
- Google's $50 clear case is priced against its own catalogue, not the market
- iOS 27 lets anyone describe a Shortcut. Someone still has to read the one it builds.
- The worst fire hour in a hyperscale build arrives before the suppression works
- Harvey builds its flagship legal model on Chinese open weights, not its investor's API
- Oura's 95% is a test result, not a sensor spec, and that is where the lawsuit lands
- Alibaba's $10.2bn placement prices AI compute off the equity market, not the profit line
- Nintendo cleared 401 GitHub repos in one day, and never went near a courtroom
Leadership14 stories
- Pick three term sheet fights, sign the rest: the clauses that outlive the round
- Six months to working AI, six years of controls: the gap in the JPMorgan cohort data
- The Arday case: a Cambridge hire made on a story nobody checked
- The $500 Billion Lease: What Boards Are Actually Financing When They Buy "AI Compute"
- The $120bn off the books: judge AI datacenter debt on tenants, not on accounting
- Meta's flattening has a due date: late April for tech, late May for business
- A $110B demand side: the first deduplicated count of what AI buyers actually pay
- If an LLM shortlists your applicants, your no-AI rule punishes the people who obey it
- Women are 26% of US AI hires and 13% of AI C-suites: the buildout is picking its bench now
- Activism at 255 campaigns: the ballot got cheap, so the activist thesis is now a standing audit
- Y'all Street's real bill: Morgan Stanley owes Dallas 3,800 seats and a hiring plan by 2027
- New York budgeted $70M for cheap groceries. It still has to find someone to run them.
- Engineering leaders are quitting with nothing lined up. Read it as a repricing, not a burnout wave
- Builders put doom at 10 to 50 per cent and expect binding rules only after the disaster
Science12 stories
- Phase III for a bespoke mRNA cancer therapy repriced the sequencing vendors, not just Moderna
- China writes off Chang'e-7's whole 2026 window, and names neither a fault nor a new date
- A nickelate finally gets its gap measured, and it does not look like a cuprate
- Vienna's plan to put a trapped-ion quantum computer inside an electron microscope
- Three lab disclosures, one control failure: the AI hacking stories are eval sandbox stories
- A nine-month balloon, and the end of the oak-leaf defence for single-use latex
- Thursday's lunar eclipse covers 96% of the moon; two of its 5.5 hours are near-invisible
- A 91-task benchmark for agent knowledge work runs every task cold, and says so
- Frontier scores at $2/$6: Grok 4.6 ties GPT-5.6 on one evaluator's index for a fifth the output price
- Chinchorro mummies took 1.4 kg of manganese a body, and left nothing to explain themselves
- A whole pot in a pit: the Florida find that is evidence about deposition, not decoration
- Curiosity's centimetre mud cracks are the only ground truth on a planet mapped in metres