OpenAI: agents found 948 Linux kernel flaws and used them to gain root privileges EN
Approved
Writing and review are carried out in separate model calls.
Review and omissions
These records concern the exact text versions in this issue. Article reviews are not a separate quality audit of the issue.
Omitted and unfinished (336)
- Kehittäjä: tekoälystartupissa pull requesteja 11 kertaa enemmän kuin aiemmin PostHogissa · EN: This article is not available in English in this issue.
- Kehittäjä: tekoälystartupissa pull requesteja 11 kertaa enemmän kuin aiemmin PostHogissa · SV: This article is not available in Swedish in this issue.
- Apple rolls out beta update with Apple Intelligence for Chinese mainland market, receiving mixed responses - Global Times: Left out in the newsroom's selection.
- iOS 27.3 Beta Brings Apple Intelligence to China With Alibaba’s Qwen AI - The Mac Observer: Left out in the newsroom's selection.
- Apple Intelligence Qwen in China: reported iOS plan - NeoTeo: Left out in the newsroom's selection.
- Apple Begins China Trial Of AI Tools Powered By Alibaba's Qwen - Dataconomy: Left out in the newsroom's selection.
- Apple begins AI trial in China, powered by Alibaba - Euronews.com: Left out in the newsroom's selection.
- Apple begins AI features rollout in China with Alibaba partnership - electronics.economictimes.indiatimes.com: Left out in the newsroom's selection.
- Revolut drops Claude in favor of a Chinese self-hosted model, cuts coding costs by 90% - Cybernews: Left out in the newsroom's selection.
- Revolut ditches Claude in favor of Chinese AI model and cuts development costs by 90% - dev.ua: Left out in the newsroom's selection.
- Revolut swaps Claude Code for a Chinese open-weight model, cutting its AI coding bill by 90% - Dealroom: Left out in the newsroom's selection.
- Can Mistral Large 4 find a critical vulnerability, prove it and fix it? We tested it on a recent Django CVE. 133 agents map the code, hunt for flaws, test attack scenarios and confirm the bug. Then…: Left out in the newsroom's selection.
- Mistral Says Mistral Large 4 Coordinated 133 Agents in Django Security Demo - www.tokenpost.com: Left out in the newsroom's selection.
- Demo Django 133-Agen Mistral Large 4: Apa yang Dibuktikannya - OrcaRouter: Left out in the newsroom's selection.
- Mistral Large 4 deploys 133 agents to find and patch a Django vulnerability - Crypto Briefing: Left out in the newsroom's selection.
- We’re sharing how our biology research team worked with Claude to map the first 3D structure of a vault particle from bacterial life. Vault particles are a longstanding mystery. They are…: Left out in the newsroom's selection.
- Investigating vault particles with claude: Left out in the newsroom's selection.
- Anthropic maps first 3D structure of a prokaryotic vault particle using Claude - Crypto Briefing: Left out in the newsroom's selection.
- Claude Helps Map a Phage Vault Particle in Anthropic’s Biology Lab - Kingy AI: Left out in the newsroom's selection.
- Read the full preprint here: https://t.co/endRhOCTko https://t.co/C2TO3Z4sAS: Left out in the newsroom's selection.
- In AA-PersonalityBench, GLM-5.3 uses the most em dashes of any model measured, averaging 7.4 per response, while Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5 use almost none We count the…: Left out in the newsroom's selection.
- Build your own personality ranking and explore more detailed results on the AA-PersonalityBench page. Explore the results: https://t.co/CVJuRfB5bJ Methodology: https://t.co/CGfYvAve9t: Left out in the newsroom's selection.
Reviewed versions (91)
Open review records by language
EN (30)
Anthropic: updating Claude Code’s subagent status line now costs about 99 per cent less EN
Approved
Writing and review are carried out in separate model calls.
Moonshot releases Kimi K3.1 Pro model weights for download EN
Approved
Writing and review are carried out in separate model calls.
US midterm voters approve dozens of local data centre restrictions EN
Approved
Writing and review are carried out in separate model calls.
ChatGPT now makes GIF animations with OpenAI’s Images 2.6 image model EN
Approved
Writing and review are carried out in separate model calls.
Ricursive says Ripple-65B beats larger US open-weight models on a coding benchmark EN
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis now measures model speed by time per task EN
Approved
Writing and review are carried out in separate model calls.
Cognition releases SWE-2 Flash coding model at what it says is one-fifth of SWE-2’s price EN
Approved
Writing and review are carried out in separate model calls.
llama.cpp 0.9.0 adds FP8 quantisation and a more efficient KV cache for multiple sequences EN
Approved
Writing and review are carried out in separate model calls.
Prime Intellect: some public benchmarks are nearly saturated after fixes EN
Approved
Writing and review are carried out in separate model calls.
OpenAI to buy Pierre Computer, whose tools Codex already uses EN
Approved
Writing and review are carried out in separate model calls.
Claude Code product leader Cat Wu moves from Anthropic to OpenAI EN
Approved
Writing and review are carried out in separate model calls.
Meta buys Conductor, whose app lets developers run several coding agents in parallel EN
Approved
Writing and review are carried out in separate model calls.
Docker Sandboxes flaw lets a sandbox redirect copies to host files EN
Approved
Writing and review are carried out in separate model calls.
Apple cut some engineers’ daily AI spending cap from $300 to $200, Bloomberg reports EN
Approved
Writing and review are carried out in separate model calls.
OpenClaw beta adds Knowledge Scribes workflows for capturing agent knowledge EN
Approved
Writing and review are carried out in separate model calls.
Nous Research brings Hermes Agent plugins and skills together in one marketplace EN
Approved
Writing and review are carried out in separate model calls.
Check Point: Haiou jailbreak bundle for Codex adds an advert to the agent’s replies EN
Approved
Writing and review are carried out in separate model calls.
Developers can try GitHub Copilot in Xcode’s built-in coding assistant in the Xcode 27.3 beta EN
Approved
Writing and review are carried out in separate model calls.
Baseten brings agent sandboxes and a model router to one platform EN
Approved
Writing and review are carried out in separate model calls.
HumanLayer, which has written about how to work with coding agents, says it is joining Factory EN
Approved
Writing and review are carried out in separate model calls.
Tencent releases Hy4 language model and promises weights in the coming days EN
Approved
Writing and review are carried out in separate model calls.
FT: Oracle and Tencent drop GPU lease deal worth an estimated $7 billion EN
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Chinese companies expect about 1 per cent of their RTX Pro 5500 orders within four months EN
Approved
Writing and review are carried out in separate model calls.
Gemini Notebook API in preview: manage notebooks and sources from code EN
Approved
Writing and review are carried out in separate model calls.
Ramp: its post-trained Qwen model beat Opus 4.6 on Ramp’s financial question task EN
Approved
Writing and review are carried out in separate model calls.
Clients of accounting firms that use TaxDome can ask Claude or ChatGPT about their affairs EN
Approved
Writing and review are carried out in separate model calls.
Harvey plans to bring Jus Mundi’s arbitration agent to joint customers this year EN
Approved
Writing and review are carried out in separate model calls.
Anthropic makes its first news licensing deal, with AP, Adweek reports EN
Approved
Writing and review are carried out in separate model calls.
Coval’s 30-day measurement: Gemini 3.5 Transcribe Live sixth for accuracy, 23rd for latency EN
Approved
Writing and review are carried out in separate model calls.
FI (31)
OpenAI: agentit löysivät 948 Linux-ytimen aukkoa ja saivat niillä pääkäyttäjän oikeudet FI
Approved
Writing and review are carried out in separate model calls.
Anthropic: Claude Coden aliagenttien tilarivin päivitys maksaa nyt noin 99 % vähemmän FI
Approved
Writing and review are carried out in separate model calls.
Moonshot julkaisi Kimi K3.1 Pro -mallin painot ladattaviksi FI
Approved
Writing and review are carried out in separate model calls.
Yhdysvaltain välivaaleissa hyväksyttiin kymmeniä paikallisia datakeskusrajoituksia FI
Approved
Writing and review are carried out in separate model calls.
ChatGPT tekee nyt GIF-animaatioita OpenAI:n Images 2.6 -kuvamallilla FI
Approved
Writing and review are carried out in separate model calls.
Ricursive: Ripple-65B ohittaa koodaustestissä isommat USA:n avointen painojen mallit FI
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis mittaa mallien nopeutta nyt tehtävään kuluvalla ajalla FI
Approved
Writing and review are carried out in separate model calls.
Cognition julkaisi SWE-2 Flash -koodausmallin: yhtiön mukaan viidennes SWE-2:n hinnasta FI
Approved
Writing and review are carried out in separate model calls.
llama.cpp 0.9.0: FP8-kvantisointi ja tehokkaampi KV-välimuisti usealle sekvenssille FI
Approved
Writing and review are carried out in separate model calls.
Prime Intellect: korjausten jälkeen osa julkisista vertailutesteistä on lähes kyllästynyt FI
Approved
Writing and review are carried out in separate model calls.
OpenAI ostaa Pierre Computerin, jonka työkaluja Codex jo käyttää FI
Approved
Writing and review are carried out in separate model calls.
Claude Coden tuotevastaava Cat Wu siirtyy Anthropicilta OpenAI:lle FI
Approved
Writing and review are carried out in separate model calls.
Meta osti Conductorin, jolla ohjelmoija ajaa useita koodausagentteja rinnakkain FI
Approved
Writing and review are carried out in separate model calls.
Docker Sandboxesin aukko: hiekkalaatikko voi ohjata kopioinnin isäntäkoneen tiedostoihin FI
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Apple laski osan insinöörien tekoälykulujen päivärajan 300 dollarista 200:aan FI
Approved
Writing and review are carried out in separate model calls.
OpenClaw lisäsi beetaan Knowledge Scribes -työnkulut agenttien tiedon tallentamiseen FI
Approved
Writing and review are carried out in separate model calls.
Nous Research kokosi Hermes Agentin lisäosat ja skillit yhteen markkinapaikkaan FI
Approved
Writing and review are carried out in separate model calls.
Check Point: Codexin jailbreak-paketti Haiou lisää agentin vastauksiin mainoksen FI
Approved
Writing and review are carried out in separate model calls.
GitHub Copilotia voi kokeilla Xcoden omassa koodausavustajassa Xcode 27.3 -beetassa FI
Approved
Writing and review are carried out in separate model calls.
Kehittäjä: tekoälystartupissa pull requesteja 11 kertaa enemmän kuin aiemmin PostHogissa FI
Approved
Writing and review are carried out in separate model calls.
Baseten tuo agenteille hiekkalaatikot ja mallireitittimen samalle alustalle FI
Approved
Writing and review are carried out in separate model calls.
Koodausagenttien työtavoista kirjoittanut HumanLayer kertoo liittyvänsä Factoryyn FI
Approved
Writing and review are carried out in separate model calls.
Tencent julkaisi Hy4-kielimallin ja lupaa painot lähipäivinä FI
Approved
Writing and review are carried out in separate model calls.
FT: Oracle ja Tencent luopuivat arviolta 7 mrd. dollarin GPU-vuokrasopimuksesta FI
Approved
Writing and review are carried out in separate model calls.
Bloomberg: kiinalaisyhtiöt odottavat neljässä kuussa noin 1 % RTX Pro 5500 -tilauksistaan FI
Approved
Writing and review are carried out in separate model calls.
Gemini Notebookin rajapinta esiversiona: muistikirjat ja lähteet hallintaan koodista FI
Approved
Writing and review are carried out in separate model calls.
Ramp: oma jälkikoulutettu Qwen-malli ohitti Opus 4.6:n Rampin talouskysymystehtävässä FI
Approved
Writing and review are carried out in separate model calls.
TaxDomea käyttävän tilitoimiston asiakas voi kysyä asioistaan Claudelta tai ChatGPT:ltä FI
Approved
Writing and review are carried out in separate model calls.
Harvey aikoo tuoda Jus Mundin välimiesmenettelyagentin yhteisasiakkailleen tänä vuonna FI
Approved
Writing and review are carried out in separate model calls.
Adweek: Anthropic teki ensimmäisen uutislisenssisopimuksensa AP:n kanssa FI
Approved
Writing and review are carried out in separate model calls.
Covalin 30 päivän mittaus: Gemini 3.5 Transcribe Live tarkkuudessa kuudes, viiveessä 23:s FI
Approved
Writing and review are carried out in separate model calls.
SV (30)
OpenAI: Agenter hittade 948 sårbarheter i Linuxkärnan och fick administratörsbehörighet genom dem SV
Approved
Writing and review are carried out in separate model calls.
Anthropic: Statusraden för subagenter i Claude Code kostar nu omkring 99 procent mindre att uppdatera SV
Approved
Writing and review are carried out in separate model calls.
Moonshot släpper modellvikterna för Kimi K3.1 Pro SV
Approved
Writing and review are carried out in separate model calls.
Dussintals lokala begränsningar av datacenter godkändes i USA:s mellanårsval SV
Approved
Writing and review are carried out in separate model calls.
ChatGPT gör nu GIF-animationer med OpenAI:s bildmodell Images 2.6 SV
Approved
Writing and review are carried out in separate model calls.
Ricursive: Ripple-65B slår större amerikanska modeller med öppna vikter i kodningstest SV
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis mäter nu modellernas hastighet med tid per uppgift SV
Approved
Writing and review are carried out in separate model calls.
Cognition släpper kodningsmodellen SWE-2 Flash: Enligt företaget en femtedel av priset för SWE-2 SV
Approved
Writing and review are carried out in separate model calls.
llama.cpp 0.9.0: FP8-kvantisering och effektivare KV-cache för flera sekvenser SV
Approved
Writing and review are carried out in separate model calls.
Prime Intellect: Efter korrigeringar är vissa offentliga benchmarkar nästan mättade SV
Approved
Writing and review are carried out in separate model calls.
OpenAI köper Pierre Computer, vars verktyg Codex redan använder SV
Approved
Writing and review are carried out in separate model calls.
Cat Wu, produktansvarig för Claude Code, går från Anthropic till OpenAI SV
Approved
Writing and review are carried out in separate model calls.
Meta har köpt Conductor, som låter programmerare köra flera kodningsagenter parallellt SV
Approved
Writing and review are carried out in separate model calls.
Sårbarhet i Docker Sandboxes: En sandlåda kan styra kopiering till värddatorns filer SV
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Apple sänkte dagsgränsen för AI-kostnader från 300 till 200 dollar för en del ingenjörer SV
Approved
Writing and review are carried out in separate model calls.
OpenClaw lade till Knowledge Scribes-arbetsflöden i beta för lagring av agenters kunskap SV
Approved
Writing and review are carried out in separate model calls.
Nous Research har samlat tillägg och skills för Hermes Agent på en marknadsplats SV
Approved
Writing and review are carried out in separate model calls.
Check Point: Jailbreakpaketet Haiou för Codex lägger in reklam i agentens svar SV
Approved
Writing and review are carried out in separate model calls.
GitHub Copilot går att prova i Xcodes egen kodningsassistent i Xcode 27.3-betan SV
Approved
Writing and review are carried out in separate model calls.
Baseten samlar sandlådor för agenter och en modellrouter på samma plattform SV
Approved
Writing and review are carried out in separate model calls.
HumanLayer, som har skrivit om arbetssätt med kodningsagenter, uppger att det ansluter sig till Factory SV
Approved
Writing and review are carried out in separate model calls.
Tencent släppte språkmodellen Hy4 och lovar modellvikterna inom de närmaste dagarna SV
Approved
Writing and review are carried out in separate model calls.
FT: Oracle och Tencent avstod från GPU-hyresavtal värt uppskattningsvis 7 miljarder dollar SV
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Kinesiska företag väntar sig cirka 1 procent av sina RTX Pro 5500-beställningar inom fyra månader SV
Approved
Writing and review are carried out in separate model calls.
Gemini Notebook API i förhandsversion: Anteckningsböcker och källor hanteras med kod SV
Approved
Writing and review are carried out in separate model calls.
Ramp: Egen eftertränad Qwen-modell gick om Opus 4.6 i Ramps uppgift med finansfrågor SV
Approved
Writing and review are carried out in separate model calls.
Kunder hos redovisningsbyråer som använder TaxDome kan fråga Claude eller ChatGPT om sina ärenden SV
Approved
Writing and review are carried out in separate model calls.
Harvey planerar att ge gemensamma kunder tillgång till Jus Mundis agent för skiljeförfarande i år SV
Approved
Writing and review are carried out in separate model calls.
Adweek: Anthropic slöt sitt första licensavtal för nyheter med AP SV
Approved
Writing and review are carried out in separate model calls.
Covals 30-dagarsmätning: Gemini 3.5 Transcribe Live sjätte i noggrannhet, 23:e i fördröjning SV
Approved
Writing and review are carried out in separate model calls.