Google opens Gemini 4 Carbon to paying developers at an introductory price 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 (380)
- Coval’s November 10 measurement of Reson8 Rumble-2 reports 2.5% word error rate and 126 ms mean time to final segment, versus 2.6% and 141 ms in its November 7 measurement.: Left out in the newsroom's selection.
- Three ways to run a model: Left out in the newsroom's selection.
- HyperWhisper benchmarks: Left out in the newsroom's selection.
- KeyASRBench 发布：用关键词错误率衡量 ASR 专名保留能力｜AI热报: Left out in the newsroom's selection.
- 160 毫秒、U=1 設定下平均錯誤率 5.76%，韓語 FLEURS 為 7.52% - TokenPost: Left out in the newsroom's selection.
- SmallVoice：Mac 菜单栏本地听写工具，178MB 语音 AI 模型跑出 88 倍速｜觉醒AI知识库: Left out in the newsroom's selection.
- 零GPU本地语音识别神器parakeet-primeline-gguf：600M模型碾压Whisper的完整指南: Left out in the newsroom's selection.
- 笔记本纯离线语音转文字：Parakeet CTC 0.6B GGUF 469MB 轻量部署实战: Left out in the newsroom's selection.
- 如何5分钟跑通 Parakeet CTC 0.6B GGUF？transcribe.cpp 语音转文字新手教程: Left out in the newsroom's selection.
- 语音识别领域的 llama.cpp - 今日头条: Left out in the newsroom's selection.
- Launch of Public Beta for Lexitas Case Intelligence™: Left out in the newsroom's selection.
- nexus-agent-framework: Left out in the newsroom's selection.
- xminal: Left out in the newsroom's selection.
- Hermes Agent: Left out in the newsroom's selection.
- Hermes Agent: Left out in the newsroom's selection.
- Manus blog: Left out in the newsroom's selection.
- Claude Code Docs – overview: Left out in the newsroom's selection.
- MCPhantom: Model Knowledge as an Unpatchable Software Supply Chain: Left out in the newsroom's selection.
- Cloudsmith launches AI agent skills registry for supply chain security - TechGig: Left out in the newsroom's selection.
- How AI security failures are endangering your supply chain – and what you can do: Left out in the newsroom's selection.
Reviewed versions (96)
Open review records by language
EN (32)
Anthropic starts IPO roadshow with a price range of $70–78 per share EN
Approved
Writing and review are carried out in separate model calls.
ChatGPT Business plans now work in Notion, Warp and Hermes Agent EN
Approved
Writing and review are carried out in separate model calls.
Anthropic adds spreadsheets, whiteboards and partner apps to Claude EN
Approved
Writing and review are carried out in separate model calls.
Google’s MKA benchmark: Claude Fable 5.2 leads, but its gap to the next two is not significant EN
Approved
Writing and review are carried out in separate model calls.
With reasoning on, Naive-N0.5-Flash matched DeepSeek in Latent Space’s app tests EN
Approved
Writing and review are carried out in separate model calls.
Anthropic gives Pro, Max, Ultra and Team subscribers one usage limit reset EN
Approved
Writing and review are carried out in separate model calls.
Anthropic: Fable 5.2 merges repeated code more often if told the code is its own EN
Approved
Writing and review are carried out in separate model calls.
Claroc: more than 35,000 Grok Build apps were open to anyone EN
Approved
Writing and review are carried out in separate model calls.
Zenity ran thousands of agent skills in isolation and says it found 100 malicious ones EN
Approved
Writing and review are carried out in separate model calls.
OpenClaw 2026.11.3 keeps private subagent results out of streamed replies EN
Approved
Writing and review are carried out in separate model calls.
Railway estimates $740,000 in unauthorised spend on stolen AI keys EN
Approved
Writing and review are carried out in separate model calls.
Adobe plans to open Photoshop and Premiere to outside AI agents EN
Approved
Writing and review are carried out in separate model calls.
Transformers adds decision models that score answer options without generating text EN
Approved
Writing and review are carried out in separate model calls.
Chrome, Edge and Firefox teams steer coding agents towards new web features EN
Approved
Writing and review are carried out in separate model calls.
Nvidia donates OpenShell, which limits what AI agents can do, to the CNCF EN
Approved
Writing and review are carried out in separate model calls.
Moonshot AI gained access to Nvidia’s top chips despite US ban, Bloomberg reports EN
Approved
Writing and review are carried out in separate model calls.
MiniMax says its H3-Image model makes a 16-megapixel image in under 10 seconds EN
Approved
Writing and review are carried out in separate model calls.
DeepMind says AI agents proved security properties of Signal’s encryption code EN
Approved
Writing and review are carried out in separate model calls.
Microsoft researchers hid a backdoor in three models by editing fewer than ten neurons EN
Approved
Writing and review are carried out in separate model calls.
OpenAI and Upwork: AI-related Upwork jobs have larger budgets and more hours EN
Approved
Writing and review are carried out in separate model calls.
OpenAI lets businesses manage their information in ChatGPT, with agents first in the US EN
Approved
Writing and review are carried out in separate model calls.
Canopy makes its Coworker agent available to accounting firms as an experimental feature EN
Approved
Writing and review are carried out in separate model calls.
Vals AI test: 23 AI models explained the law better than they pointed people to local help EN
Approved
Writing and review are carried out in separate model calls.
Law firm Crosby takes out $5 million liability insurance for its AI agents EN
Approved
Writing and review are carried out in separate model calls.
Nvidia shareholders file class action over Jensen Huang’s relationship with suspected spy EN
Approved
Writing and review are carried out in separate model calls.
AI companies pledge shutdown mechanisms in Paris declaration – with no sanctions EN
Approved
Writing and review are carried out in separate model calls.
Australia plans incident-reporting duty for AI companies after OpenAI agent breach EN
Approved
Writing and review are carried out in separate model calls.
The Register: Microsoft to switch on Claude by default in UK Copilot on 9 December EN
Approved
Writing and review are carried out in separate model calls.
New York’s AI safety law takes effect in January as registration opens for large developers EN
Approved
Writing and review are carried out in separate model calls.
ElevenLabs’ own test: Scribe v3 makes 27 per cent fewer word errors than Scribe v2 EN
Approved
Writing and review are carried out in separate model calls.
Otter adds 27 transcription languages – Finnish coming soon, company says EN
Approved
Writing and review are carried out in separate model calls.
FI (32)
Google avasi Gemini 4 Carbonin maksaville kehittäjille esittelyhinnalla FI
Approved
Writing and review are carried out in separate model calls.
Anthropic aloitti listautumiskiertueen: osakkeen hintahaarukka 70–78 dollaria FI
Approved
Writing and review are carried out in separate model calls.
ChatGPT Business -tilausta voi nyt käyttää Notionissa, Warpissa ja Hermes Agentissa FI
Approved
Writing and review are carried out in separate model calls.
Anthropic tuo Claudeen taulukot ja valkotaulut sekä kumppanien sovellukset FI
Approved
Writing and review are carried out in separate model calls.
Googlen MKA-testi: Claude Fable 5.2 kärjessä, ero kahteen seuraavaan ei merkitsevä FI
Approved
Writing and review are carried out in separate model calls.
Naive-N0.5-Flash pysyi päättely päällä DeepSeekin tasolla Latent Spacen sovellustesteissä FI
Approved
Writing and review are carried out in separate model calls.
Anthropic antaa Pro-, Max-, Ultra- ja Team-tilaajille yhden käyttörajojen nollauksen FI
Approved
Writing and review are carried out in separate model calls.
Anthropic: Fable 5.2 yhdistää toistuvaa koodia useammin, jos koodi kerrotaan sen omaksi FI
Approved
Writing and review are carried out in separate model calls.
Claroc: yli 35 000 Grok Build -sovellusta oli kenen tahansa avattavissa FI
Approved
Writing and review are carried out in separate model calls.
Zenity ajoi tuhansia agenttiskillejä eristetysti ja kertoo löytäneensä 100 haitallista FI
Approved
Writing and review are carried out in separate model calls.
OpenClaw 2026.11.3 pitää yksityiset aliagenttitulokset poissa striimatuista vastauksista FI
Approved
Writing and review are carried out in separate model calls.
Railway: varastetuilla tekoälyavaimilla luvattomia kuluja arviolta 740 000 dollaria FI
Approved
Writing and review are carried out in separate model calls.
Adobe aikoo avata Photoshopin ja Premieren ulkopuolisille tekoälyagenteille FI
Approved
Writing and review are carried out in separate model calls.
Päätösmallit Transformersiin: vastausvaihtoehdot pisteytetään tuottamatta tekstiä FI
Approved
Writing and review are carried out in separate model calls.
Chromen, Edgen ja Firefoxin tiimit ohjaavat koodausagentteja uusiin web-ominaisuuksiin FI
Approved
Writing and review are carried out in separate model calls.
Nvidia lahjoitti tekoälyagenttien toimia rajaavan OpenShellin CNCF-säätiölle FI
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Moonshot AI sai Nvidian huippusiruja käyttöönsä USA:n kiellosta huolimatta FI
Approved
Writing and review are carried out in separate model calls.
MiniMaxin H3-Image-kuvamalli tekee yhtiön mukaan 16 megapikselin kuvan alle 10 sekunnissa FI
Approved
Writing and review are carried out in separate model calls.
DeepMind: tekoälyagentit todistivat Signalin salauskoodin turvaominaisuuksia FI
Approved
Writing and review are carried out in separate model calls.
Microsoftin tutkijat kätkivät kolmeen malliin takaoven muokkaamalla alle kymmentä neuronia FI
Approved
Writing and review are carried out in separate model calls.
OpenAI ja Upwork: tekoälyyn liittyvissä Upwork-töissä isommat budjetit ja enemmän tunteja FI
Approved
Writing and review are carried out in separate model calls.
OpenAI antaa yritysten hallita tietojaan ChatGPT:ssä – omat agentit ensin Yhdysvalloissa FI
Approved
Writing and review are carried out in separate model calls.
Canopy avasi Coworker-agentin tilitoimistoasiakkailleen kokeellisena ominaisuutena FI
Approved
Writing and review are carried out in separate model calls.
Vals AI:n testi: 23 tekoälymallia selitti lakia paremmin kuin ohjasi paikalliseen apuun FI
Approved
Writing and review are carried out in separate model calls.
Asianajotoimisto Crosby otti 5 miljoonan dollarin vastuuvakuutuksen tekoälyagenteilleen FI
Approved
Writing and review are carried out in separate model calls.
Nvidia-osakkaat nostivat ryhmäkanteen Jensen Huangin suhteesta vakoojaksi epäiltyyn FI
Approved
Writing and review are carried out in separate model calls.
Tekoälyyhtiöt lupaavat Pariisin julistuksessa sammutusmekanismit – sanktioita ei ole FI
Approved
Writing and review are carried out in separate model calls.
Australia valmistelee tekoälyyhtiöille ilmoitusvelvollisuutta OpenAI-agentin murron vuoksi FI
Approved
Writing and review are carried out in separate model calls.
The Register: Microsoft kytkee Clauden oletuksena päälle Britannian Copilotissa 9.12. FI
Approved
Writing and review are carried out in separate model calls.
New Yorkin tekoälylaki voimaan tammikuussa – suurten kehittäjien rekisteröinti alkoi FI
Approved
Writing and review are carried out in separate model calls.
ElevenLabsin oma testi: Scribe v3 tekee 27 % vähemmän sanavirheitä kuin Scribe v2 FI
Approved
Writing and review are carried out in separate model calls.
Otter lisää litterointiin 27 kieltä – suomi yhtiön mukaan tulossa pian FI
Approved
Writing and review are carried out in separate model calls.
SV (32)
Google öppnade Gemini 4 Carbon för betalande utvecklare till introduktionspris SV
Approved
Writing and review are carried out in separate model calls.
Anthropic har inlett roadshow inför börsnoteringen – prisintervall 70–78 dollar per aktie SV
Approved
Writing and review are carried out in separate model calls.
ChatGPT Business-abonnemang kan nu användas i Notion, Warp och Hermes Agent SV
Approved
Writing and review are carried out in separate model calls.
Anthropic lägger till kalkylark, whiteboards och partnerappar i Claude SV
Approved
Writing and review are carried out in separate model calls.
Googles MKA-benchmark: Claude Fable 5.2 i topp, skillnaden mot de två närmaste inte signifikant SV
Approved
Writing and review are carried out in separate model calls.
Naive-N0.5-Flash höll DeepSeeks nivå med resonemang påslaget i Latent Spaces apptester SV
Approved
Writing and review are carried out in separate model calls.
Anthropic ger Pro-, Max-, Ultra- och Team-abonnenter en återställning av användningsgränserna SV
Approved
Writing and review are carried out in separate model calls.
Anthropic: Fable 5.2 slår oftare ihop upprepad kod om koden sägs vara modellens egen SV
Approved
Writing and review are carried out in separate model calls.
Claroc: Över 35 000 Grok Build-appar kunde öppnas av vem som helst SV
Approved
Writing and review are carried out in separate model calls.
Zenity körde tusentals agentskills isolerat och uppger sig ha hittat 100 skadliga SV
Approved
Writing and review are carried out in separate model calls.
OpenClaw 2026.11.3 håller privata subagentresultat borta från strömmade svar SV
Approved
Writing and review are carried out in separate model calls.
Railway: Stulna AI-nycklar gav obehöriga kostnader på uppskattningsvis 740 000 dollar SV
Approved
Writing and review are carried out in separate model calls.
Adobe planerar att öppna Photoshop och Premiere för externa AI-agenter SV
Approved
Writing and review are carried out in separate model calls.
Beslutsmodeller i Transformers: Svarsalternativ poängsätts utan att text genereras SV
Approved
Writing and review are carried out in separate model calls.
Teamen bakom Chrome, Edge och Firefox vägleder kodningsagenter till nya webbfunktioner SV
Approved
Writing and review are carried out in separate model calls.
Nvidia skänkte OpenShell, som begränsar AI-agenters handlingar, till CNCF SV
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Moonshot AI fick tillgång till Nvidias toppchipp trots amerikanskt förbud SV
Approved
Writing and review are carried out in separate model calls.
Bildmodellen H3-Image från MiniMax skapar enligt företaget en bild på 16 megapixlar på under 10 sekunder SV
Approved
Writing and review are carried out in separate model calls.
DeepMind: AI-agenter bevisade säkerhetsegenskaper i Signals krypteringskod SV
Approved
Writing and review are carried out in separate model calls.
Microsofts forskare gömde en bakdörr i tre modeller genom att ändra färre än tio neuroner SV
Approved
Writing and review are carried out in separate model calls.
OpenAI och Upwork: Större budgetar och fler timmar i AI-relaterade Upwork-uppdrag SV
Approved
Writing and review are carried out in separate model calls.
OpenAI låter företag hantera sina uppgifter i ChatGPT – egna agenter först i USA SV
Approved
Writing and review are carried out in separate model calls.
Canopy öppnar AI-agenten Coworker för sina byråkunder som experimentell funktion SV
Approved
Writing and review are carried out in separate model calls.
Vals AI:s test: 23 AI-modeller förklarade lagen bättre än de hänvisade till lokal hjälp SV
Approved
Writing and review are carried out in separate model calls.
Advokatbyrån Crosby tecknade ansvarsförsäkring på 5 miljoner dollar för sina AI-agenter SV
Approved
Writing and review are carried out in separate model calls.
Nvidias aktieägare väcker grupptalan om Jensen Huangs relation till misstänkt spion SV
Approved
Writing and review are carried out in separate model calls.
AI-företag lovar avstängningsmekanismer i Parisdeklarationen – sanktioner saknas SV
Approved
Writing and review are carried out in separate model calls.
Australien förbereder rapporteringsplikt för AI-företag efter intrång av en OpenAI-agent SV
Approved
Writing and review are carried out in separate model calls.
The Register: Microsoft aktiverar Claude som standard i Copilot i Storbritannien 9 december SV
Approved
Writing and review are carried out in separate model calls.
New Yorks AI-lag träder i kraft i januari – registreringen av stora utvecklare har börjat SV
Approved
Writing and review are carried out in separate model calls.
ElevenLabs eget test: Scribe v3 gör 27 procent färre ordfel än Scribe v2 SV
Approved
Writing and review are carried out in separate model calls.
Otter lägger till 27 transkriberingsspråk – finska kommer snart enligt företaget SV
Approved
Writing and review are carried out in separate model calls.