OpenAI opens GPT-6 Cyber security model to selected Daybreak Red customers 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 (468)
- No Result, No Charge: Outcome-Based Pricing and the Future of AI Monetization for Finance: Left out in the newsroom's selection.
- Tunnistammeko tekoälyn riskit?: Left out in the newsroom's selection.
- Usage metrics API adds code review details and addressed rate: Left out in the newsroom's selection.
- Usage metrics API flags telemetry and version gaps: Left out in the newsroom's selection.
- Secret scanning upgrades detectors, adds SpaceX as partner: Left out in the newsroom's selection.
- New pull request overview page in public preview: Left out in the newsroom's selection.
- RT @depoll: Confession: I’ve scrolled past dozens of comments on a PR just to find out why I couldn’t merge it. More times than I’d like to…: Left out in the newsroom's selection.
- China Merchants' LionRock AI Lab Open-Sources Lion-ER-35B-A3B Embodied Reasoning Model: Left out in the newsroom's selection.
- Moxin Technology Launches MoWorld 2.0, an Ascend-Native 4D World Model That Simulates Physics: Left out in the newsroom's selection.
- Huawei's Xiaoyi Work Is First to Run openPangu-2.1-Pro, Billed for HarmonyOS Task Enhancement: Left out in the newsroom's selection.
- AI 에이전트를 위한 멀티모달 의사결정 모델 ‘Kanana-2-v-decide’ 오픈소스 공개: Left out in the newsroom's selection.
- Every technological revolution has remodeled our institutions – how might AI? In @ScienceMagazine, Microsoft researcher @glenweyl and co-author @profjamesevans explore ‘reverse alignment,’ the idea…: Left out in the newsroom's selection.
- How do you solve the trust problem with AI-generated code? AI agents can generate code at a scale too great for humans to review line-by-line. That makes it difficult to ensure the code will do…: Left out in the newsroom's selection.
- RT @GitHubNext: How we can secure 50% of the world's software. Traditionally, formal methods have been mostly used for safety-critical sys…: Left out in the newsroom's selection.
- So many agents So many features This thread is going to be 100 posts long soon: Left out in the newsroom's selection.
- RT @particle_api: Our token efficient model particle-1 is now in OpenClaw https://t.co/TVVA5eemF0: Left out in the newsroom's selection.
- Best Lawyers’ 2027 U.S. law-firm rankings report that 28% of firms with 20 or more lawyers fully implemented generative AI tools across multiple practice areas, compared with 11% in the previous year’: Left out in the newsroom's selection.
- QX Global Group launched QX TaxReady in the United States, combining agentic AI with human experts to help CPA firms manage tax-preparation workloads. The service uses human support for cases or softw: Left out in the newsroom's selection.
- CLM Provider Leah Launches Self-Service AI Tool-Building Feature: Left out in the newsroom's selection.
- CLM Provider Leah Launches Self-Service AI Tool-Building Feature - Law.com: Left out in the newsroom's selection.
Reviewed versions (102)
Open review records by language
EN (34)
METR: Anthropic’s models are improving about twice as fast on well-specified tasks EN
Approved
Writing and review are carried out in separate model calls.
US chip stock index falls 12 per cent as investors sell AI shares EN
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis: Inkling 2 Small matches Gemini 3.8 Flash on its index EN
Approved
Writing and review are carried out in separate model calls.
Google’s Gemini Omni 1.2 Flash makes 15-second videos – price falls 30 per cent EN
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis ranks FLUX 3 Image Klein second among open-weight models for image generation EN
Approved
Writing and review are carried out in separate model calls.
Visa’s own benchmark: AI models failed 42.8 per cent of shopping task runs EN
Approved
Writing and review are carried out in separate model calls.
Codex CLI 0.167.0 marks unread results in the agent command center EN
Approved
Writing and review are carried out in separate model calls.
Hermes Agent 0.23.0 lets users run the desktop app in a browser from their own gateway EN
Approved
Writing and review are carried out in separate model calls.
OpenClaw Multiplayer lets several people steer the same agent with shared context EN
Approved
Writing and review are carried out in separate model calls.
Check Point: ChatGPT’s Lockdown Mode could be bypassed through HTML preview EN
Approved
Writing and review are carried out in separate model calls.
Mistral adds managed agents and apps to Studio, with an option to run them on-premises EN
Approved
Writing and review are carried out in separate model calls.
Reuters: about 600,000 Nvidia H200 chips await China’s import approval in Hong Kong EN
Approved
Writing and review are carried out in separate model calls.
Moonshot offers businesses a harness for Kimi agents as a cloud service EN
Approved
Writing and review are carried out in separate model calls.
DeepSeek says Harness has 1.54 million daily users through its API EN
Approved
Writing and review are carried out in separate model calls.
Baidu releases ERNIE 5.5 Flash base model weights for further training EN
Approved
Writing and review are carried out in separate model calls.
China bars SoftBank from buying ABB’s robotics business in China EN
Approved
Writing and review are carried out in separate model calls.
CNBC: Chinese lab Moonshot sought Nvidia Blackwell compute in European data centres EN
Approved
Writing and review are carried out in separate model calls.
China requires AI agents preinstalled on phones to be tested for network access licences EN
Approved
Writing and review are carried out in separate model calls.
SuperCLUE benchmark: Gemini-4-Argon scores highest, Kimi-K3.1-Pro leads Chinese models EN
Approved
Writing and review are carried out in separate model calls.
Harness changed models’ compliance with banking rules – effect varied by model EN
Approved
Writing and review are carried out in separate model calls.
Small-model agents sometimes wrote unsupported authorisations and restrictions into their summaries EN
Approved
Writing and review are carried out in separate model calls.
OpenAI: GPT-6 Astra performed worse on tasks framed as safety research EN
Approved
Writing and review are carried out in separate model calls.
False pre-screening note overturned up to 40 per cent of agent judges’ correct verdicts EN
Approved
Writing and review are carried out in separate model calls.
Box says agent search built on Nvidia models could cut token use by 90 per cent EN
Approved
Writing and review are carried out in separate model calls.
Harvey unveils fact graph for disputes, linking facts to their sources EN
Approved
Writing and review are carried out in separate model calls.
On a podcast, Opzer demonstrates a Claude skill that checks tax returns for discrepancies EN
Approved
Writing and review are carried out in separate model calls.
Sage plans to open an accounts payable AI Workspace to selected Intacct customers in November EN
Approved
Writing and review are carried out in separate model calls.
Claude starts reading bank account data as Personal Finance enters beta in the US EN
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Microsoft tightens oversight of some neocloud data centre projects EN
Approved
Writing and review are carried out in separate model calls.
OpenAI pledges not to require NDAs from officials for its own data centres EN
Approved
Writing and review are carried out in separate model calls.
Microsoft promises to bring Copilot’s Claude models within the EU Data Boundary by the end of the year EN
Approved
Writing and review are carried out in separate model calls.
OpenAI’s new mode stores API requests selected for safety review in the customer’s cloud EN
Approved
Writing and review are carried out in separate model calls.
Epoch AI estimates compute in Broadcom’s TPU shipments doubled in a quarter EN
Approved
Writing and review are carried out in separate model calls.
FI (34)
OpenAI avasi GPT-6 Cyber -tietoturvamallin valituille Daybreak Red -asiakkaille FI
Approved
Writing and review are carried out in separate model calls.
METR: Anthropicin mallit paranevat tarkasti määritellyissä tehtävissä noin tuplavauhtia FI
Approved
Writing and review are carried out in separate model calls.
Yhdysvaltain siruosakeindeksi laski 12 %, kun sijoittajat myivät tekoälyosakkeita FI
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis: Inkling 2 Small yltää indeksissä Gemini 3.8 Flashin tasolle FI
Approved
Writing and review are carried out in separate model calls.
Googlen Gemini Omni 1.2 Flash tekee 15 sekunnin videoita – hinta laski 30 % FI
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis: FLUX 3 Image Klein toisena avointen painojen malleista kuvanluonnissa FI
Approved
Writing and review are carried out in separate model calls.
Visan oma vertailutesti: tekoälymallit epäonnistuivat 42,8 %:ssa ostotehtävien ajoista FI
Approved
Writing and review are carried out in separate model calls.
Codex CLI 0.167.0: lukemattomat tulokset näkyvät agenttien komentokeskuksessa FI
Approved
Writing and review are carried out in separate model calls.
Hermes Agent 0.23.0: työpöytäsovellusta voi käyttää selaimessa omalta yhdyskäytävältä FI
Approved
Writing and review are carried out in separate model calls.
OpenClaw Multiplayer: usea ihminen ohjaa samaa agenttia yhteisellä kontekstilla FI
Approved
Writing and review are carried out in separate model calls.
Check Point: ChatGPT:n Lockdown-tilan pystyi kiertämään HTML-esikatselun kautta FI
Approved
Writing and review are carried out in separate model calls.
Mistral Studioon hallitut agentit ja sovellukset – ajettavissa myös omilla palvelimilla FI
Approved
Writing and review are carried out in separate model calls.
Reuters: noin 600 000 Nvidian H200-sirua odottaa Hongkongissa Kiinan tuontilupaa FI
Approved
Writing and review are carried out in separate model calls.
Moonshot tarjoaa yrityksille Kimi-agenttien harnessia pilvipalveluna FI
Approved
Writing and review are carried out in separate model calls.
DeepSeekin mukaan Harnessilla on API:n kautta 1,54 miljoonaa päivittäistä käyttäjää FI
Approved
Writing and review are carried out in separate model calls.
Baidu julkaisi ERNIE 5.5 Flash -pohjamallin painot jatkokoulutusta varten FI
Approved
Writing and review are carried out in separate model calls.
Kiina kielsi SoftBankia ostamasta ABB:n robotiikkaliiketoimintaa Kiinassa FI
Approved
Writing and review are carried out in separate model calls.
CNBC: Kiinalainen Moonshot haki Nvidian Blackwell-laskentaa Euroopan datakeskuksista FI
Approved
Writing and review are carried out in separate model calls.
Kiina vaatii testaamaan puhelimiin esiasennetut tekoälyagentit verkkolupamenettelyssä FI
Approved
Writing and review are carried out in separate model calls.
SuperCLUEn testi: Gemini-4-Argon sai eniten pisteitä, Kimi-K3.1-Pro paras kiinalainen FI
Approved
Writing and review are carried out in separate model calls.
Harness muutti mallien pankkisääntöjen noudattamista – vaikutus vaihteli malleittain FI
Approved
Writing and review are carried out in separate model calls.
Pienten mallien agenttien tiivistelmissä oli joskus perusteettomia lupia ja rajoituksia FI
Approved
Writing and review are carried out in separate model calls.
OpenAI: GPT-6 Astra suoriutui turvallisuustutkimukseksi kuvatuista tehtävistä heikommin FI
Approved
Writing and review are carried out in separate model calls.
Väärä esitarkastusmerkintä kumosi jopa 40 % arvioija-agenttien oikeista arvioista FI
Approved
Writing and review are carried out in separate model calls.
Box: Nvidian malleihin perustuva agenttihaku voi vähentää tokenien käyttöä 90 % FI
Approved
Writing and review are carried out in separate model calls.
Harvey esitteli riita-asioihin fact graphin, joka kytkee tosiasiat lähteisiinsä FI
Approved
Writing and review are carried out in separate model calls.
Opzer esittelee podcastissa Claude-skillin, joka etsii veroilmoitusten ristiriitoja FI
Approved
Writing and review are carried out in separate model calls.
Sage aikoo avata ostolaskujen tekoälytyötilan valituille Intacct-asiakkaille marraskuussa FI
Approved
Writing and review are carried out in separate model calls.
Claude alkaa lukea pankkitilien tietoja – Personal Finance beetana Yhdysvalloissa FI
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Microsoft valvoo osaa neopilvien datakeskushankkeista tiukemmin FI
Approved
Writing and review are carried out in separate model calls.
OpenAI lupaa olla vaatimatta päättäjiltä salassapitosopimuksia omissa datakeskuksissaan FI
Approved
Writing and review are carried out in separate model calls.
Microsoft lupaa Copilotin Claude-mallit EU:n tietorajan piiriin vuoden loppuun mennessä FI
Approved
Writing and review are carried out in separate model calls.
OpenAI:n uudessa tilassa turvatarkastettavat API-pyynnöt tallennetaan asiakkaan pilveen FI
Approved
Writing and review are carried out in separate model calls.
Epoch AI arvioi: Broadcomin TPU-toimitusten laskentateho kaksinkertaistui neljänneksessä FI
Approved
Writing and review are carried out in separate model calls.
SV (34)
OpenAI öppnade säkerhetsmodellen GPT-6 Cyber för utvalda Daybreak Red-kunder SV
Approved
Writing and review are carried out in separate model calls.
METR: Anthropics modeller förbättras ungefär dubbelt så snabbt på väldefinierade uppgifter SV
Approved
Writing and review are carried out in separate model calls.
USA:s halvledarindex föll 12 procent när investerare sålde AI-aktier SV
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis: Inkling 2 Small når samma nivå som Gemini 3.8 Flash i indexet SV
Approved
Writing and review are carried out in separate model calls.
Googles Gemini Omni 1.2 Flash gör 15 sekunder långa videor – priset sjönk med 30 procent SV
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis: FLUX 3 Image Klein tvåa bland modeller med öppna vikter inom bildgenerering SV
Approved
Writing and review are carried out in separate model calls.
Visas eget benchmark: AI-modeller misslyckades i 42,8 procent av köpkörningarna SV
Approved
Writing and review are carried out in separate model calls.
Codex CLI 0.167.0: Olästa resultat visas i kommandocentralen för agenter SV
Approved
Writing and review are carried out in separate model calls.
Hermes Agent 0.23.0: Skrivbordsappen kan användas i webbläsaren via egen gateway SV
Approved
Writing and review are carried out in separate model calls.
OpenClaw Multiplayer: Flera personer styr samma agent med gemensam kontext SV
Approved
Writing and review are carried out in separate model calls.
Check Point: ChatGPT:s Lockdown Mode kunde kringgås via HTML-förhandsvisningen SV
Approved
Writing and review are carried out in separate model calls.
Mistral Studio får hanterade agenter och appar – kan även köras på egna servrar SV
Approved
Writing and review are carried out in separate model calls.
Reuters: Omkring 600 000 H200-chip från Nvidia väntar i Hongkong på kinesiskt importtillstånd SV
Approved
Writing and review are carried out in separate model calls.
Moonshot erbjuder företag en harness för Kimi-agenter som molntjänst SV
Approved
Writing and review are carried out in separate model calls.
Enligt DeepSeek har Harness 1,54 miljoner dagliga användare via API:t SV
Approved
Writing and review are carried out in separate model calls.
Baidu publicerar modellvikterna till ERNIE 5.5 Flash Base för vidareträning SV
Approved
Writing and review are carried out in separate model calls.
Kina förbjöd SoftBank att köpa ABB:s robotikverksamhet i Kina SV
Approved
Writing and review are carried out in separate model calls.
CNBC: Kinesiska Moonshot sökte Nvidias Blackwell-kapacitet i europeiska datacenter SV
Approved
Writing and review are carried out in separate model calls.
Kina kräver att förinstallerade AI-agenter i telefoner testas vid tillståndsprövningen för nätanslutning SV
Approved
Writing and review are carried out in separate model calls.
SuperCLUE:s test: Gemini-4-Argon fick flest poäng, Kimi-K3.1-Pro bäst av de kinesiska SV
Approved
Writing and review are carried out in separate model calls.
Harnessen ändrade hur modeller följde bankregler – effekten varierade mellan modellerna SV
Approved
Writing and review are carried out in separate model calls.
Sammanfattningar från agenter med små modeller innehöll ibland ogrundade tillstånd och begränsningar SV
Approved
Writing and review are carried out in separate model calls.
OpenAI: GPT-6 Astra presterade sämre på uppgifter som beskrevs som säkerhetsforskning SV
Approved
Writing and review are carried out in separate model calls.
Falsk förhandsgranskningsnotering kullkastade upp till 40 procent av bedömaragenters korrekta bedömningar SV
Approved
Writing and review are carried out in separate model calls.
Box: Agentsökning byggd på Nvidias modeller kan minska tokenanvändningen med 90 procent SV
Approved
Writing and review are carried out in separate model calls.
Harvey presenterade fact graph för tvister – kopplar fakta till deras källor SV
Approved
Writing and review are carried out in separate model calls.
Opzer visar i en podcast en Claude-skill som letar efter motstridigheter i skattedeklarationer SV
Approved
Writing and review are carried out in separate model calls.
Sage planerar att i november öppna en AI-arbetsyta för leverantörsfakturor för utvalda Intacct-kunder SV
Approved
Writing and review are carried out in separate model calls.
Claude börjar läsa uppgifter från bankkonton – Personal Finance i beta i USA SV
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Microsoft skärper tillsynen över vissa datacenterprojekt hos neocloudbolag SV
Approved
Writing and review are carried out in separate model calls.
OpenAI lovar att inte kräva sekretessavtal av beslutsfattare vid sina egna datacenter SV
Approved
Writing and review are carried out in separate model calls.
Microsoft lovar att Claude-modellerna i Copilot hamnar innanför EU-datagränsen före årets slut SV
Approved
Writing and review are carried out in separate model calls.
I OpenAI:s nya läge sparas API-anrop i kundens moln inför säkerhetsgranskning SV
Approved
Writing and review are carried out in separate model calls.
Epoch AI uppskattar: Beräkningskapaciteten i Broadcoms TPU-leveranser fördubblades på ett kvartal SV
Approved
Writing and review are carried out in separate model calls.