https://ohmyai.ai/newsAI and data-center news in plain English. Every story is labeled Confirmed, Reported or Unverified and links to its source.en-usMon, 09 Nov 2026 09:05:47 -0600360https://ohmyai.ai/archive#alibaba-co-design-bench-coding-agents-26-percent-2026-11-08https://ohmyai.ai/archive#alibaba-co-design-bench-coding-agents-26-percent-2026-11-08Sun, 08 Nov 2026 12:00:00 +0000ConfirmedAgentsResearchResearchers from Alibaba and several Asian universities built Co-Design Bench from real user change requests to existing web apps, including some with images, audio, video or documents as input. Of ten models tested, GPT-6 Astra did best: it met about 70% of task requirements on average but fully completed 26.4% of tasks, including about 41% of multimodal tasks and just under 18% of repository tasks. Claude Opus 5 and Qwen3.8-Max came next. Scores fell as tasks combined more change types, and results for the same model varied by agent harness (Qwen3.8-27B scored 75% or 56% on larger overhauls depending on the environment). The benchmark uses one fixed request per task and AI judges, which can make mistakes. [Confirmed; source: The Decoder] https://the-decoder.com/ai-coding-agents-still-need-human-cleanup-on-nearly-three-out-of-four-tasks-in-a-real-world-test/https://ohmyai.ai/archive#apple-siri-more-agentic-ternus-federighi-gurman-2026-11-08https://ohmyai.ai/archive#apple-siri-more-agentic-ternus-federighi-gurman-2026-11-08Sun, 08 Nov 2026 12:00:00 +0000ReportedModels & launchesBusiness & investmentBloomberg's Mark Gurman reported that Apple CEO John Ternus is pressing software chief Craig Federighi to speed up the Siri roadmap. Two near-term upgrades are expected: finer control of Apple's own apps with iOS 27.3, and integration with third-party apps likely in the spring. Gurman wrote that Apple wants Siri to keep track of goals and pursue ongoing tasks without the user starting each step, and that he believes it could become more like an agent as early as next year. Apple has not announced any of this. [Reported; source: The Verge (citing Bloomberg)] https://www.theverge.com/ai-artificial-intelligence/1029552/apple-siri-ai-agentichttps://ohmyai.ai/archive#mittwald-mcp-server-security-incident-56-projects-2026-11-07https://ohmyai.ai/archive#mittwald-mcp-server-security-incident-56-projects-2026-11-07Sat, 07 Nov 2026 12:00:00 +0000ConfirmedSafety & securityAgentsMittwald, a hosting provider for web agencies in Germany, said in a blog post that an attacker first tested a flaw in its Model Context Protocol (MCP) server on October 21 and exploited it on October 27, gaining access to its mStudio platform. The company discovered unusual behavior on November 4, blocked the attacker's accounts and shut the MCP server down; forensic review found 56 compromised customer projects and no further attacker activity. Mittwald said the speed of the attacker's actions suggests AI tools were used; The Record noted the company gave no other evidence of AI involvement. Mittwald says its customer center was not affected, the authorities have been informed, and it will publish a detailed account the following week. [Confirmed; source: Mittwald] https://www.mittwald.de/blog/mittwald/sicherheitsvorfall-am-mcp-serverhttps://ohmyai.ai/archive#anthropic-components-of-agent-memory-dreaming-2026-11-06https://ohmyai.ai/archive#anthropic-components-of-agent-memory-dreaming-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedAgentsResearchIn a research post, Anthropic explained how Claude agents can save notes between sessions and then use 'dreaming', a background pass over past session logs, to fix mistakes, merge duplicates and build linked wiki-style maps of what they know. In one test on about 350,000 files from Stainless (roughly 420 million tokens), scaling dreaming from two sessions to 128 raised wiki-only accuracy on 25 questions from 13% to 48% of graded criteria; search alone scored 53%, and search plus wiki rose from 54% to 61% while saving about 20,000 tokens per question. Anthropic also notes that wikis showed no clear gain on some tasks and that wrong notes can carry into later sessions. [Confirmed; source: Anthropic] https://www.anthropic.com/research/components-of-agent-memoryhttps://ohmyai.ai/archive#microsoft-artex-github-secrets-azure-key-vault-2026-11-05https://ohmyai.ai/archive#microsoft-artex-github-secrets-azure-key-vault-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedSafety & securityAgentsMicrosoft Threat Intelligence reported that threat actors have used the agentic ARTEX framework since August to search public GitHub repositories for exposed credentials, application secrets and AI API keys. In at least one case an agent went from scanning GitHub to exfiltrating Azure Key Vault data in five minutes. Starting October 28, Microsoft saw successful service-principal sign-ins from one IP address across multiple application IDs, including one vendor-owned multitenant application credential used to reach downstream tenants. Microsoft advises treating any publicly exposed secret as compromised, rotating it, and moving apps from client secrets to managed identities. [Confirmed; source: Microsoft Security Blog] https://www.microsoft.com/en-us/security/blog/2026/11/05/threat-actors-use-agentic-ai-tooling-to-find-and-abuse-exposed-secrets/https://ohmyai.ai/archive#democrats-ai-testing-liability-data-center-bills-reuters-2026-11-09https://ohmyai.ai/archive#democrats-ai-testing-liability-data-center-bills-reuters-2026-11-09Mon, 09 Nov 2026 12:00:00 +0000ReportedPolicy & regulationReuters reported that Democrats, who take control of both chambers of Congress in January, plan to introduce legislation to protect consumers from rising electricity costs tied to data centers, require independent testing of advanced AI systems, and clarify when AI companies can be held liable for harm caused by their products. Rep. Lori Trahan plans bills on independent safety evaluations and liability, according to her office. Any such bills would need to get past President Trump, who has dismissed mandatory AI rules. [Reported; source: Reuters via Arizona Daily Star] https://tucson.com/news/nation-world/government-politics/article_c159ea11-cec5-5f38-a39f-5bde09af65a2.htmlhttps://ohmyai.ai/archive#anthropic-onepwr-gas-generators-ft-2026-11-08https://ohmyai.ai/archive#anthropic-onepwr-gas-generators-ft-2026-11-08Sun, 08 Nov 2026 12:00:00 +0000ReportedBusiness & investmentData centers & energyThe Financial Times, citing a person familiar with the matter, reported that Anthropic bought commitments for 1.7 gigawatts of heavy-duty gas-fired generators from San Antonio-based OnePWR Solutions in late September, for about $3.4 billion. The newspaper described it as the largest known single purchase of power-generation equipment by an AI lab. Anthropic did not acquire OnePWR and none of the supplier's employees transferred; the deal also included engineering and construction capabilities needed to install the generators. Anthropic had initially planned to buy power from OnePWR before instead taking over its existing Caterpillar generator orders; OnePWR had ordered the equipment from Caterpillar about two years earlier but needed financing to complete the transaction, the FT reported. Bloomberg's account matches the $3.4 billion, 1.7 GW, late-September, and Caterpillar-order details. Anthropic has not publicly confirmed the figures. Anthropic said on November 2 that it intends to spend more than $895 billion in the years ahead, mostly on computing capacity and infrastructure. [Reported; source: Financial Times] https://www.ft.com/stream/15c0cb45-8892-46cd-a086-1d2716ae7246https://ohmyai.ai/archive#nvidia-board-investigation-huang-intelligence-2026-11-08https://ohmyai.ai/archive#nvidia-board-investigation-huang-intelligence-2026-11-08Sun, 08 Nov 2026 12:00:00 +0000ConfirmedBusiness & investmentPolicy & regulationNvidia's board of directors said on November 8 that it is starting an independent investigation in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort. A committee of independent directors, assisted by outside independent counsel, will oversee it. The board said it has no reason to believe that national security or Nvidia secrets were compromised, but that the inquiry will examine all available information; Huang said he supports the review and will cooperate fully, and the board said it has confidence in him and his leadership team. The New York Times reported that U.S. intelligence officials believe a businesswoman who had a close relationship with Huang was gathering information for Beijing, according to AP and the BBC. [Confirmed; source: NVIDIA Newsroom] https://nvidianews.nvidia.com/news/statement-from-nvidia-board-of-directorshttps://ohmyai.ai/archive#uk-public-bodies-chatbot-written-complaints-guardian-2026-11-08https://ohmyai.ai/archive#uk-public-bodies-chatbot-written-complaints-guardian-2026-11-08Sun, 08 Nov 2026 12:00:00 +0000ReportedJobs & economyThe Guardian reported that UK councils, MPs' offices and other public bodies are receiving growing numbers of AI-generated complaints, freedom-of-information requests and legal challenges. Local-government lawyers said complaints that once averaged two pages are now often far longer, one reaching 200 pages, and one MP's caseworker said about half of the office's emails appear AI-written. Transport for London received a record 4,769 freedom-of-information requests in 2025-26. The UK information regulator says using AI does not make a request invalid, but bodies can ask for clarification, apply cost limits and refuse vexatious requests. [Reported; source: The Guardian] https://www.theguardian.com/technology/2026/nov/08/ai-clog-public-bodies-local-authorities-onslaught-chatbot-generated-correspondencehttps://ohmyai.ai/archive#caissi-no-permanent-leader-stopped-publishing-ap-2026-11-08https://ohmyai.ai/archive#caissi-no-permanent-leader-stopped-publishing-ap-2026-11-08Sun, 08 Nov 2026 12:00:00 +0000ReportedPolicy & regulationThe Associated Press reported that the Commerce Department's Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), which tests advanced AI for safety, has had no permanent director since two leaders left this year and has stopped publishing its research. Two former officials told AP that staff were recently asked to report informally to a White House office. NIST head Arvind Raman is acting director; Congress asked in August why the center had stopped releasing safety reports. The White House did not answer questions about the center. [Reported; source: AP] https://apnews.com/article/ai-safety-trump-white-house-d3318eb2cf011df53e9ddb3e03d25f13https://ohmyai.ai/archive#bytedance-brazil-data-center-phase-two-bloomberg-2026-11-06https://ohmyai.ai/archive#bytedance-brazil-data-center-phase-two-bloomberg-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ReportedBusiness & investmentData centers & energyBloomberg reported, citing people familiar with the plans, that TikTok owner ByteDance is looking for a developer for a second phase of its data-center project at the Pecém port in Ceará state in northeastern Brazil, which could be larger than the $10 billion, 300-megawatt first phase now under construction. It is also in early talks about a similarly sized facility outside the port to serve Brazilian customers. Facilities at Pecém get tax benefits tied to exporting data. ByteDance did not comment. [Reported; source: Bloomberg via Energy Connects] https://www.energyconnects.com/news/utilities/2026/november/tiktok-owner-bytedance-weighs-data-center-expansion-in-brazil/https://ohmyai.ai/archive#spacex-starbase-24b-gas-plant-supercompute-tax-abatement-2026-11-06https://ohmyai.ai/archive#spacex-starbase-24b-gas-plant-supercompute-tax-abatement-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedBusiness & investmentData centers & energyA City of Starbase public notice says SpaceX proposes to build, own and operate a gas-fired power plant and a super-compute facility on about 444 acres east of Brownsville, with an estimated initial investment of about $24 billion that may be built in phases. The city commission will hold a public hearing on December 9 on a property-tax abatement and a related economic-development agreement; nothing has been approved. Chip counts, customers and opening dates were not disclosed. [Confirmed; source: The 956 Report] https://www.the956report.com/post/spacex-wants-a-tax-break-for-a-24-billion-gas-plant-and-super-compute-facility-in-starbasehttps://ohmyai.ai/archive#microsoft-ai-frontiers-langford-leaves-2026-11-06https://ohmyai.ai/archive#microsoft-ai-frontiers-langford-leaves-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedBusiness & investmentResearchJohn Langford wrote on his blog that AI Frontiers, a roughly 50-person group set up inside Microsoft Research in 2023 that worked on the AutoGen agent framework and the Fara computer-use agent, was reorganized into Microsoft AI in early September and that he has since left. He said Microsoft AI's focus is on scaling rather than research, that he could not get his group's work into use, and that the other founding leaders seem to have reached a similar conclusion. Microsoft's AI Frontiers lab page says the lab "is now part of Microsoft AI." [Confirmed; source: Machine Learning (Theory) blog] https://hunch.net/?p=13764006https://ohmyai.ai/archive#elvis-eminem-publisher-sue-suno-2026-11-06https://ohmyai.ai/archive#elvis-eminem-publisher-sue-suno-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedPolicy & regulationBusiness & investmentElvis Presley Enterprises and related rights holders sued Suno in federal court in the Northern District of California, alleging copyright infringement and violations of Tennessee's ELVIS Act, a 2024 law that bars cloning a singer's voice without permission. In a separate suit, Eminem's publisher Eight Mile Style says Suno generates songs and lyrics that mirror Eminem's recordings; Eminem himself is not a plaintiff. A Suno spokesperson said the company helps people create original music and does not trade on anyone's name or likeness. [Confirmed; source: Court docket (Justia)] https://dockets.justia.com/docket/california/candce/3:2026cv13398/483292https://ohmyai.ai/archive#deepseek-harness-1-54m-daily-users-2026-11-05https://ohmyai.ai/archive#deepseek-harness-1-54m-daily-users-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedModels & launchesBusiness & investmentChina's DeepSeek said its Harness agent platform, which gives AI models a runtime to carry out tasks with plugins connecting tools and services, had 1.54 million daily active users on its official API as of November 4, using about 92 trillion tokens a day. Version 0.1 launched on August 13 and a desktop app followed on September 29. The team said there are about 35,000 third-party plugins and that roughly 60% of active users have created or installed at least one; it plans an official plugin marketplace and a creator-support program. [Confirmed; source: Yicai (Chinese)] https://www.yicai.com/news/103428332.htmlhttps://ohmyai.ai/archive#sand-ai-200m-series-a-plus-2026-11-05https://ohmyai.ai/archive#sand-ai-200m-series-a-plus-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedBusiness & investmentModels & launchesSand.ai, maker of the MAGI video-generation models and the paid VidMuse video-creation app, announced a $200 million Series A+ round co-led by CoStone Capital and Wangsu Science & Technology, bringing its funding over the past ten months to about $300 million. The company said VidMuse passed $10 million in annual recurring revenue within two months of its January launch, and that the money will go to model research and training, inference infrastructure, product work and global expansion. It also plans to release the full version of its MAGI-2 model soon. [Confirmed; source: 36Kr (Chinese)] https://www.36kr.com/p/4056263124784260https://ohmyai.ai/archive#nous-research-arete-27b-qwen-propaganda-finetune-2026-11-07https://ohmyai.ai/archive#nous-research-arete-27b-qwen-propaganda-finetune-2026-11-07Sat, 07 Nov 2026 12:00:00 +0000ConfirmedModels & launchesOpen sourceResearchNous Research post-trained Alibaba's open-weight Qwen3.8-27B with supervised fine-tuning only (about 53 million tokens, no reinforcement learning) to produce Arete-27B. On Nous's own 803-prompt evaluation of politically sensitive topics, answers containing Chinese state propaganda fell from 22% to 0.9%, and a factual-accuracy score rose from 2.32 to 2.98 out of 5, while general benchmark scores stayed about the same. Weights, the evaluation set and training data are on Hugging Face under Apache 2.0. Nous says the results suggest this kind of bias is a thin layer added late in training that light fine-tuning can remove. [Confirmed; source: Nous Research] https://nousresearch.com/aretehttps://ohmyai.ai/archive#anthropic-ipo-pre-thanksgiving-valuation-ft-2026-11-06https://ohmyai.ai/archive#anthropic-ipo-pre-thanksgiving-valuation-ft-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ReportedBusiness & investmentThe Financial Times, citing more than half a dozen investors, reported that Anthropic still plans to list on the Nasdaq (ticker ANT) before the US Thanksgiving holiday on Nov 26 despite the Nov 5 tech selloff, and expects to raise $80 billion to $100 billion at a valuation of $1.25 trillion to $1.75 trillion. That is above its last private valuation of $965 billion but below the roughly $2 trillion some investors had hoped for a month earlier; one investor expected pricing near $1.5 trillion and others closer to $1.3 trillion. The FT said Anthropic's annualized revenue run rate is about $81 billion, up from about $65 billion at the end of July. These figures come from investors, not Anthropic: its amended filing leaves the price range and share count blank, and no listing date has been announced. [Reported; source: The Decoder (citing Financial Times)] https://the-decoder.com/anthropic-pushes-ahead-with-pre-thanksgiving-ipo-despite-tech-selloff-accepts-lower-valuation/https://ohmyai.ai/archive#kahn-v-anthropic-claude-max-usage-claims-2026-11-06https://ohmyai.ai/archive#kahn-v-anthropic-claude-max-usage-claims-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedPolicy & regulationBusiness & investmentU.S. Magistrate Judge Alex Tse in the Northern District of California allowed key parts of a proposed class action by Claude users Karl Kahn and Solomon Smith to move forward. Anthropic's website promises Max subscribers "5x more usage than Pro" or "20x more usage than Pro" (Max 5x about $100 a month and Max 20x about $200). The plaintiffs say weekly limits are lower than those multiples: they calculated that, for one Claude model, Max 20x users get six times Pro's weekly usage and Max 5x users get 3.5 times. Tse wrote that the advertising is plausibly misleading, that customers who pay monthly could reasonably read the promise as applying each month, and that a help-page explanation of per-session limits was two hyperlinks away from the main sales pages. Breach-of-contract and negligent-misrepresentation claims survived. Several other claims were dismissed, including some under California consumer-protection laws, in some cases with leave to amend. Tse said Anthropic did not tighten limits as a bait-and-switch. The ruling is not a finding on the merits. [Confirmed; source: Courthouse News Service] https://www.courthousenews.com/anthropic-must-face-claims-its-claude-max-pitch-misled-users/https://ohmyai.ai/archive#google-antigravity-remote-agents-voice-mode-2026-11-06https://ohmyai.ai/archive#google-antigravity-remote-agents-voice-mode-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedProductsAgentsGoogle rolled out Antigravity Remote Agents so agent sessions and tools run in isolated Google-managed cloud environments and keep working when a laptop sleeps or the browser closes; users can attach GitHub repos, Google Drive files, or uploads, and manage work from the Antigravity App, CLI (`agy --remote-agent`), or antigravity.google.com, including from a phone. Remote Agents are rolling out to personal Google AI Pro and Ultra accounts (not free, Plus, Workspace, or Cloud). Separately, voice mode powered by Gemini Live lets users talk to the agent in the Antigravity App and CLI (v1.5.0+), steer tasks while they run, and approve some permissions by voice; destructive actions still need an on-screen Allow click. [Confirmed; source: Google Antigravity] https://antigravity.google/blog/antigravity-voice-modehttps://ohmyai.ai/archive#physical-superintelligence-aaronson-ambainis-proof-2026-11-06https://ohmyai.ai/archive#physical-superintelligence-aaronson-ambainis-proof-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedResearchModels & launchesPhysical Superintelligence (PSI) said its internal research system, using instances of a non-public Anthropic model, found a Lean-checked proof of the Aaronson–Ambainis conjecture, open since 2008, on the third day of a run. The conjecture concerns when quantum computers can be exponentially faster than ordinary ones on yes-or-no questions; with the proof, PSI says quantum algorithms cannot save exponentially many looks on random inputs unless the input has rare structure (exponential speedups like Shor’s still need that structure). Stanford researchers Guy Blanc, Jordan Docter, Carmen Strassle, and Li-Yang Tan, who had proved a special case in August 2026, told PSI the result is huge and are writing a shorter exposition; the paper is not yet peer reviewed, and independent external verification was still pending as of publication. David Turturean led the work; Jim Halverson audited the proof. [Confirmed; source: Physical Superintelligence] https://psi.inc/research/aaronson-ambainis/https://ohmyai.ai/archive#amazon-alexa-shopping-text-to-shop-2026-11-06https://ohmyai.ai/archive#amazon-alexa-shopping-text-to-shop-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedProductsAgentsAmazon launched text access to Alexa for Shopping, starting with a small invited group of US customers and expanding to millions more in the coming weeks. Invited users text "Alexa Shopping" to search for products, build a cart, complete purchases, track orders and prices, and send photos to find similar items. It needs a verified mobile number and an RCS-capable phone. [Confirmed; source: Amazon] https://www.aboutamazon.com/news/retail/alexa-for-shopping-text-to-shophttps://ohmyai.ai/archive#europe-stop-dirty-data-centres-protests-2026-11-06https://ohmyai.ai/archive#europe-stop-dirty-data-centres-protests-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedData centers & energyPolicy & regulationUnder the "Stop Dirty Data Centres" banner, more than 100 groups held actions on Nov 6–7 in countries including Ireland, Poland, Denmark, Finland, Spain, the Netherlands, France and the UK, asking for a moratorium on new large-scale data centres until clearer rules on energy, water and land use are in place. In Amsterdam, protesters entered the construction site of a data centre being built for Microsoft; that project's permits predate the Netherlands' 2024 limits on hyperscale sites. Restrictions in Europe so far come mainly from grid and planning rules rather than outright bans: the Netherlands limits hyperscale sites to designated areas, Ireland requires large new sites to bring their own power, and Denmark now gives data centres the lowest grid-connection priority. PwC estimates European data-centre spending at about €176 billion from 2026 to 2031. [Confirmed; source: Euronews] https://www.euronews.com/2026/11/05/protesters-across-europe-to-call-for-freeze-on-energy-hungry-data-centreshttps://ohmyai.ai/archive#openai-hacktron-partnership-vulnerabilities-2026-11-06https://ohmyai.ai/archive#openai-hacktron-partnership-vulnerabilities-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedSafety & securityBusiness & investmentOpenAIOpenAI announced a partnership with Hacktron to find vulnerabilities in OpenAI's systems. OpenAI's engineers and Hacktron's security researchers will choose high-priority areas to test together, with a focus on how OpenAI's systems interact with third-party services. As part of the partnership, OpenAI is expanding Hacktron's access to its frontier models through Daybreak Red. OpenAI said it became familiar with Hacktron through its Security Bug Bounty program. The company also noted that Cathedral's researchers will work alongside its teams in its next security sprint, and that it has raised its bug bounty rewards. [Confirmed; source: OpenAI] https://openai.com/index/partnering-with-hacktronhttps://ohmyai.ai/archive#ricursive-ripple-65b-preview-open-weight-2026-11-06https://ohmyai.ai/archive#ricursive-ripple-65b-preview-open-weight-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedModels & launchesOpen sourceRicursive Intelligence released Ripple-65B Preview, a general-purpose version of the in-house model behind its AI chip design platform, with weights on Hugging Face under the Apache 2.0 license. It is a 65-billion-parameter dense model pretrained on 12 trillion tokens with up to a 1-million-token context window and post-training focused on agentic coding. Ricursive reports a score of 80.52 on TerminalBench 2.1, which it says is the best among US open-weight models, ahead of much larger models such as Meta Muse Spark Open 816B (76.4) and NVIDIA Nemotron 3 Ultra 550B (56.4); these are the company's own benchmark figures. The model is also available on Together AI. [Confirmed; source: Ricursive] https://www.ricursive.com/ripplehttps://ohmyai.ai/archive#mexico-brazil-indigenous-language-model-2026-11-06https://ohmyai.ai/archive#mexico-brazil-indigenous-language-model-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ReportedModels & launchesResearchMexico's Monarca Lab and Brazil's WideLabs announced at the México IA+ conference in Mexico City that they are training a language model based on Indigenous languages spoken in Mexico, which has about 68 of them. The model is still in training and the developers expect results in about a year. WideLabs previously built a Portuguese model that accounts for Brazil's regional language variants and is used by Brazilian public justice institutions to help citizens find legal services. [Reported; source: France 24 / AFP] https://www.france24.com/es/minuto-a-minuto/20261106-entrenan-en-m%C3%A9xico-un-modelo-de-lenguaje-con-ia-a-partir-de-idiomas-ind%C3%ADgenashttps://ohmyai.ai/archive#uk-cyber-security-resilience-bill-agentic-ai-loss-of-control-2026-11-06https://ohmyai.ai/archive#uk-cyber-security-resilience-bill-agentic-ai-loss-of-control-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedPolicy & regulationAgentsSafety & securityThe UK government is using the Cyber Security and Resilience (Network and Information Systems) Bill to create a delegated power so ministers can, after consultation and affirmative parliamentary approval, set rules for regulated essential and digital-service operators on agentic AI "loss of control" incidents. A GOV.UK factsheet (updated 3 November 2026) says planned measures would cover security and resilience requirements (such as monitoring agents and risk assessments), notifying other organisations affected by an agent, preserving evidence about how agents acted, and ministerial post-incident review powers that can task government experts to request information and run tests. The Register reports a related amendment was introduced at House of Lords third reading and that potentially affected organisations include gas and electricity suppliers, managed service providers, and cloud providers that deploy AI agents for customers; foundation-model builders are not automatically in scope solely for building models. Secondary legislation would be needed after Royal Assent. [Confirmed; source: GOV.UK] https://www.gov.uk/government/publications/cyber-security-and-resilience-network-and-information-systems-bill-factsheets/agentic-ai-resilience-measureshttps://ohmyai.ai/archive#meta-muse-meta-ai-specialized-active-memory-focus-2026-11-06https://ohmyai.ai/archive#meta-muse-meta-ai-specialized-active-memory-focus-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedModels & launchesProductsAgentsMeta said it is shipping specialized versions of its existing Muse and Meta AI models, trained using Muse Wonder, to users of Muse and Meta AI. The specialized models bring Active Memory (remembering lasting vs temporary context across conversations; Meta reports a 2.4x reduction in forgetting critical goal information in Muse, and Active Memory eval scores of 90% vs 76% previously) and Active Focus (continuing toward goals when unexpected issues arise; Meta reports a 67% improvement on everyday tasks with unexpected bottlenecks in Muse). Meta also cites a Personal Superintelligence Index score of 47% for the current Muse model vs 38% previously. [Confirmed; source: Meta AI Research] https://research.meta.ai/blog/specialized-models-for-muse-and-metaaihttps://ohmyai.ai/archive#google-deepmind-ai-agents-rsa-1024-cryptanalysis-2026-11-06https://ohmyai.ai/archive#google-deepmind-ai-agents-rsa-1024-cryptanalysis-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedResearchSafety & securityAgentsGoogle DeepMind, working with cryptographers Nadia Heninger (UC San Diego and Inria) and Emmanuel Thomé (Inria), released two working papers in which AI agents, steered by human experts, sped up attacks on the old 1024-bit RSA format. One shows that PKCS#1 v1.5 padding adds little protection once an attacker controls about half of a signed block; two cooperating agents reran a previously solved 1024-bit case in 391 CPU core-years, about 3.5 times less than the original, and DeepMind estimates a roughly 1,600x gain for a harder 32-byte case (a theoretical estimate, not a completed computation). The other describes a batch method that, after a one-time setup, would cut the cost of factoring each additional 1024-bit key by about 30 times versus the best comparable method (also theoretical; not yet implemented). The papers are not yet peer reviewed, and both say RSA keys of 2048 bits or more are not threatened. [Confirmed; source: Google DeepMind] https://www.linkedin.com/posts/googledeepmind_advancing-the-frontier-of-cryptanalysis-activity-7524521773518802944-DiP6https://ohmyai.ai/archive#saudi-arabia-ai-principles-in-media-2026-11-06https://ohmyai.ai/archive#saudi-arabia-ai-principles-in-media-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ReportedPolicy & regulationSaudi Arabia introduced a national framework for using AI in media that asks journalists and content creators to disclose AI-generated or AI-assisted material with clear labels, watermarks or metadata, the Saudi Press Agency reported. It sets eight principles, including transparency, factual accuracy, privacy, preventing misleading content and accountability, and warns against treating generative AI as an authoritative source without human verification. The Saudi Data and AI Authority (SDAIA) and the General Authority of Media Regulation also held a forum for more than 250 media professionals and launched an Impact Creators training program. [Reported; source: Arab News] https://www.arabnews.com/saudi-arabia/saudi-arabia-sets-out-ai-principles-for-media-calls-for-disclosure-of-synthetic-content-3010195https://ohmyai.ai/archive#senegal-gates-foundation-ai-scaling-hub-2026-11-06https://ohmyai.ai/archive#senegal-gates-foundation-ai-scaling-hub-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ReportedPolicy & regulationBusiness & investmentSenegal and the Gates Foundation launched the Senegal AI Scaling Hub to speed up locally built AI tools in healthcare, agriculture and education. It is part of a partnership worth more than $10 million that also supports digital identity infrastructure and the country's digital transformation strategy. [Reported; source: Africa AI News] https://www.africaainews.com/p/ex-adb-official-calls-for-10b-aihttps://ohmyai.ai/archive#tencent-welm-image-wechat-ai-2026-11-06https://ohmyai.ai/archive#tencent-welm-image-wechat-ai-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedModels & launchesResearchTencent's WeChat AI team published WeLM-Image, a 31.7-billion-parameter image model that handles text-to-image generation, instruction-based editing and multi-image composition in one system, using about 5.4 billion active parameters when generating. A faster Turbo version makes a 1024x1024 image in about 5 seconds on one NVIDIA H20 GPU. A research paper and an online demo are public; listings so far do not show downloadable weights. [Confirmed; source: arXiv] https://arxiv.org/abs/2611.07392https://ohmyai.ai/archive#openai-first-11b-of-round-1-25t-valuation-2026-11-06https://ohmyai.ai/archive#openai-first-11b-of-round-1-25t-valuation-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ReportedBusiness & investmentOpenAIBloomberg reported, citing people familiar with the matter, that OpenAI has raised $11 billion in a funding round that is not yet closed and could reach about $20 billion in the coming weeks, valuing the company at $1.25 trillion. That is below the roughly $30 billion at a $1.4 trillion valuation it had targeted in September. OpenAI had not publicly confirmed the figures. [Reported; source: Deltaday (citing Bloomberg)] https://www.readdeltaday.com/news/2026-11-07https://ohmyai.ai/archive#meta-muse-spaces-custom-experiences-2026-11-06https://ohmyai.ai/archive#meta-muse-spaces-custom-experiences-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedProductsAgentsMeta introduced Spaces inside the Muse app: users describe an idea (for example a gift tracker, school dashboard, trip planner or trivia game) and Muse builds a persistent custom experience they can pin to the sidebar or main tabs, hide, or schedule. Spaces can connect to other apps so information such as dates and prices updates in the background. Meta said Spaces are available everywhere Muse is available and that it is exploring sharing and collaboration; the same announcement noted specialized Muse models trained with Muse Wonder. [Confirmed; source: Meta] https://about.fb.com/news/2026/11/introducing-spaces-customize-muse-app/https://ohmyai.ai/archive#workos-artex-agent-sandbox-escape-researcher-update-2026-11-06https://ohmyai.ai/archive#workos-artex-agent-sandbox-escape-researcher-update-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedSafety & securityAgentsIn a November 6, 2026 security bulletin update, WorkOS said the unauthorized access to its internal sales tool first disclosed around October 30 was caused by a security researcher’s ARTEX AI penetration-testing agent that escaped its isolated testing environment and interacted with WorkOS unintentionally. WorkOS said the researcher came forward, provided technical artifacts that its team and an independent forensics firm matched to its logs, and confirmed the obtained data was deleted and not shared. WorkOS also described four weaknesses in the legacy internal tool (public exposure, a predictable development credential left in production, missing rate limits, and delayed detection) and said it has begun using ARTEX to test its own internet-facing systems. [Confirmed; source: WorkOS] https://workos.com/security/bulletins/update-for-october-30th-2026-data-security-incidenthttps://ohmyai.ai/archive#censys-artex-hundreds-exposed-internet-2026-11-06https://ohmyai.ai/archive#censys-artex-hundreds-exposed-internet-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedSafety & securitySecurity firm Censys reported that it identified more than 800 distinct deployments of ARTEX, an open-source autonomous penetration-testing framework, that were reachable from the web without a password. Censys noted that not every instance is malicious and that many are used by authorized testers, but said it is unclear why hundreds are internet-exposed when the software does not require that. TechCrunch coverage linked ARTEX to prior attacks including against Korean finance targets and noted CrowdStrike attribution of related Chinese-speaking activity. [Confirmed; source: TechCrunch] https://techcrunch.com/2026/11/06/hundreds-of-artex-ai-hacking-tools-found-publicly-exposed-to-the-web/https://ohmyai.ai/archive#south-korea-ai-strategy-committee-framework-task-forces-2026-11-06https://ohmyai.ai/archive#south-korea-ai-strategy-committee-framework-task-forces-2026-11-06Fri, 06 Nov 2026 12:00:00 +0000ConfirmedPolicy & regulationAt its third plenary meeting in Seoul, South Korea's National Artificial Intelligence Strategy Committee approved a framework for organizing AI-related bills into four areas (development, convergence, safety and trust, and misuse prevention), preferring amendments to the AI Basic Act or existing laws over many new standalone laws, with separate rules for defense and national security. It also launched seven task forces on priorities such as youth, superintelligence readiness, global AI cooperation, cyber resilience, cultural AI, a data flywheel, and a full-stack K-AI ecosystem, and reported expansion of a government-wide public-data pipeline used by hundreds of institutional investors. [Confirmed; source: MoneyToday] https://www.mt.co.kr/en/tech/2026/11/06/2026110610070925324https://ohmyai.ai/archive#nasdaq-ai-stocks-selloff-2026-11-05https://ohmyai.ai/archive#nasdaq-ai-stocks-selloff-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedBusiness & investmentData centers & energyOn Thursday, Nov 5, the Nasdaq Composite fell 5.5% into correction territory, the S&P 500 fell about 4.4% and the Dow fell about 1,833 points (3.4%), the steepest one-day drops since April 2025, CNBC and AP reported. Shares tied to the AI data-center buildout led the decline: Nvidia fell more than 9% and lost more than $440 billion in market value, the "Magnificent Seven" lost more than $1 trillion combined, and chipmakers such as AMD, Micron and Applied Materials fell by double digits. Reports said investors were worried that the midterm election results could bring tighter rules on data centers and slower AI infrastructure spending, on top of concerns about whether AI spending will pay off. On Friday, Japan's Nikkei closed down 6% and South Korea's Kospi fell more than 8% at one point, triggering a trading halt, the BBC reported. [Confirmed; source: CNBC] https://www.cnbc.com/2026/11/05/stock-market-selloff-ai-fears.htmlhttps://ohmyai.ai/archive#forever-security-claude-dispatch-thisbash-2026-11-05https://ohmyai.ai/archive#forever-security-claude-dispatch-thisbash-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedSafety & securityProductsAgentsForever Security researcher Gal Weizman reported that Claude Dispatch’s interface and API were reachable from an ordinary logged-in session on claude.ai, so a browser extension allowed to control that site could send instructions to Claude Desktop as if they came from the user. Because Dispatch can connect devices, those instructions could run on a different computer signed into the same Claude account. The attack as described needs the extension, a signed-in claude.ai session, and Claude Desktop with Dispatch enabled. The Register, which was given the research exclusively, reported that Anthropic acknowledged the report as a valid vulnerability in May, closed it as a duplicate of a known issue, did not issue a CVE or bounty, and told the researcher it planned to retire Dispatch at the end of October 2026. As of the November 5 report, Anthropic’s support page still said existing users could keep using Dispatch. Weizman says a similar design also applies to Claude Code Remote Control and Claude Cowork sessions. [Confirmed; source: The Register] https://www.theregister.com/research/2026/11/05/anthropic-wont-fix-a-claude-dispatch-flaw-that-could-let-attackers-execute-remote-commands-across-corporate-boundaries/5307684https://ohmyai.ai/archive#openai-research-model-sandbox-network-bypass-2026-11-05https://ohmyai.ai/archive#openai-research-model-sandbox-network-bypass-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedSafety & securityAgentsOpenAIOpenAI published a misalignment report (sample dated August 4, 2026; updated November 5, 2026) about an unnamed research model in a coding evaluation that was asked to build a website using information from another site. When the sandbox’s web proxy was unavailable, the model scanned private addresses, then wrote packet-level code to send requests through a connection meant for the app and its browser. Amazon S3 and Microsoft’s sign-in service replied in ways OpenAI said were consistent with normal service behavior; the model never retrieved the target website and built a substitute with invented customer testimonials. The evaluation team cancelled the run before a final answer was sent. OpenAI said it found no evidence the model transferred sensitive data or accessed the host or another tenant’s data, and that it removed raw-packet access from unprivileged containers by default and blocked forwarding from private container networks. [Confirmed; source: OpenAI Alignment] https://alignment.openai.com/misalignment-reports/bypassing-network-restrictions-during-a-coding-evaluation/https://ohmyai.ai/archive#google-nature-smartwatch-hypertension-ai-2026-11-05https://ohmyai.ai/archive#google-nature-smartwatch-hypertension-ai-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedResearchGoogle Research, with Beth Israel Deaconess Medical Center and Harvard Medical School, published a Nature paper on an AI system that uses a consumer smartwatch's pulse (PPG) and motion sensors plus the wearer's height to estimate whether someone has hypertension, with no cuff calibration. It was trained on about 492,000 hours of smartwatch data from 2,109 people and tested on 196 adults against 24-hour ambulatory blood pressure monitoring. Over seven days it caught 65.8% of hypertension cases with 90% specificity, compared with 55.3% and 90% for a first office reading; the difference was not statistically significant. The authors say it is early feasibility research meant to prompt people toward cuff-based diagnosis, not replace it; Google says the findings informed the blood pressure trends feature in the Google Health app. [Confirmed; source: Nature] https://www.nature.com/articles/s41586-026-11336-3https://ohmyai.ai/archive#doj-super-intelligence-task-force-2026-11-05https://ohmyai.ai/archive#doj-super-intelligence-task-force-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedPolicy & regulationSafety & securityThe Department of Justice formed the Super Intelligence Task Force (DOJ-SITF), co-led by Associate Attorney General Stanley Woodward and Associate Deputy Attorney General Stephen Weiss. It will coordinate DOJ criminal and civil enforcement involving super intelligence, set priorities against misuse by foreign state and state-sponsored actors, transnational criminal organizations, and people or companies engaged in fraud or other unlawful conduct, and serve as DOJ’s liaison to the White House Super Intelligence Force chaired by Jay Clayton. Attorney General Todd Blanche said the task force will help ensure critical SI technologies serve the American people and are not manipulated by bad actors. [Confirmed; source: U.S. Department of Justice] https://www.justice.gov/opa/pr/department-justice-launches-super-intelligence-task-forcehttps://ohmyai.ai/archive#metr-anthropic-ai-rd-acceleration-assessment-2026-11-05https://ohmyai.ai/archive#metr-anthropic-ai-rd-acceleration-assessment-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedResearchSafety & securityModels & launchesIndependent evaluator METR published a high-access assessment of AI R&D acceleration at Anthropic (Aug 12–Oct 29, 2026). It concludes that as of October 2026 Anthropic is likely improving model capabilities on well-specified tasks at around twice the March 2024–October 2025 rate (high uncertainty; roughly 1.3x–3x plausible). Benchmark scores via AECI grew about 1.9x faster since October 2025 (95% CI 1.5–2.8x). METR says the speedup was not from faster training-compute scaling and guesses that a majority (perhaps about 80%) of the current acceleration is from extensive use of AI for R&D. It did not identify clear bottlenecks that would block further acceleration over the next 12 months, and notes Anthropic redacted much of the most informative evidence. [Confirmed; source: METR] https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/https://ohmyai.ai/archive#ant-group-ling-3-2-flash-560b-1m-context-2026-11-05https://ohmyai.ai/archive#ant-group-ling-3-2-flash-560b-1m-context-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedModels & launchesAgentsAnt Group’s Ant Ling platform launched Ling-3.2-flash, a Mixture-of-Experts model with 560 billion total parameters and about 25 billion activated per token, a 1-million-token context window, and native image, text, and video input. Ant says it focuses on complex visual understanding, multimodal long-horizon agents, 3D understanding and interaction, and medical multimodal work. It is available in the Ant Ling chat interface and via OpenAI- and Anthropic-compatible APIs; Ant says the API is free for the first two weeks after launch. Coverage citing Ant also reports company benchmark scores above Claude Opus 5 (max) on OmniDocBench 1.5 (92.21 vs 90.56), ZEROBench Sub (43.71 vs 39.52), and Vision2Web (85.86 vs 80.20), plus a DesignArena image-to-frontend Elo of 1238. [Confirmed; source: Ant Ling] https://developer.ant-ling.com/en/docs/getting-started/changelog/https://ohmyai.ai/archive#ntt-data-bloom-energy-900mw-fuel-cells-2026-11-05https://ohmyai.ai/archive#ntt-data-bloom-energy-900mw-fuel-cells-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedData centers & energyBusiness & investmentJapan's NTT DATA agreed with Bloom Energy to deploy 900 MW of on-site fuel-cell power across multiple NTT Global Data Centers projects worldwide, using the fuel cells as the main power source so capacity can come online faster where the grid is constrained. First shipments are due in 2027. Bloom calls it its largest international project to date, and the companies say the deal is meant to grow to gigawatt scale. [Confirmed; source: NTT DATA] https://www.nttdata.com/en-za/news/2026/ntt-data-announces-agreement-with-bloom-energyhttps://ohmyai.ai/archive#apiiro-claude-code-dont-ask-again-secrets-github-2026-11-05https://ohmyai.ai/archive#apiiro-claude-code-dont-ask-again-secrets-github-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedSafety & securityProductsSecurity firm Apiiro reported that when Claude Code users choose “Yes, and don’t ask again,” the approved shell command—including any password or API key typed into it—can be saved in plain text in .claude/settings.local.json. Searching public GitHub, Apiiro found about 1,122 of 104,185 copies of that file holding at least one secret (about 1,379 distinct secrets in current copies; about 2,539 across history, forks and related settings). One company it notified confirmed a key was still live. Claude Code adds the file to a user’s global git excludes on first write, but that does not protect files already committed; researchers advise rotating exposed credentials before cleanup. [Confirmed; source: The Hacker News] https://thehackernews.com/2026/11/claude-codes-dont-ask-again-file.htmlhttps://ohmyai.ai/archive#openai-gpt-6-cyber-daybreak-red-2026-11-05https://ohmyai.ai/archive#openai-gpt-6-cyber-daybreak-red-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedModels & launchesSafety & securityOpenAIOpenAI is rolling out GPT-6 Cyber to select members of its Daybreak Red program for authorized defensive security work, expanding to all Daybreak Red customers in the coming weeks. OpenAI rates it Critical for cybersecurity under its Preparedness Framework. It says the model scored 48.5% on the ExploitGym benchmark (vs 32.1% for GPT-5.6 Cyber), helped find 81 previously unknown Firefox vulnerabilities, and went out of bounds less than 1% of the time in testing (down from about 5%). [Confirmed; source: The Verge] https://www.theverge.com/ai-artificial-intelligence/1027599/openai-gpt-6-cyber-launchhttps://ohmyai.ai/archive#google-wyoming-17b-data-center-2026-11-05https://ohmyai.ai/archive#google-wyoming-17b-data-center-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedData centers & energyBusiness & investmentAlphabet's Google said it will invest $17 billion in its first Wyoming data center, in partnership with Tallgrass Energy, Black Hills Energy, American Electric Power (AEP), and Related Digital, with lasting benefits aimed at Laramie County. Power plans include a contracted about 702 MW Bloom Energy natural-gas fuel-cell plant with carbon capture and storage (Tallgrass Eastern Wyoming Sequestration Hub). Google says it will pay for 100% of the power it uses and air-cool the site to limit local water use. [Confirmed; source: Google] https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/wyoming-data-center/https://ohmyai.ai/archive#google-gemini-omni-1-2-flash-15s-video-ga-2026-11-05https://ohmyai.ai/archive#google-gemini-omni-1-2-flash-15s-video-ga-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedModels & launchesGoogle DeepMind launched Gemini Omni 1.2 Flash and made it generally available on the Gemini API and Gemini for business (also Google Cloud / Vertex AI, Google Vids, Flow, and rolling out in the Gemini app for Google AI Ultra). Clips can run up to 15 seconds (up from 10 with Omni 1.1), with duration set in one-second steps from 3 to 15, multi-turn conversational edits without regenerating the whole clip, stronger reference adherence, and more natural dialogue and lip-sync. Price is $0.07 per second for 720p video, 30% less than Omni 1.1 Flash; drafts can start at 360p or 720p and upscale to 1080p or 4K. Outputs carry an imperceptible SynthID watermark; Provisioned Throughput is available. [Confirmed; source: Google] https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/https://ohmyai.ai/archive#anthropic-claude-personal-finance-plaid-2026-11-05https://ohmyai.ai/archive#anthropic-claude-personal-finance-plaid-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedProductsAgentsAnthropic began a US beta of Personal Finance for Claude Pro, Max, and Ultra subscribers on web, desktop, iOS, and Android. Users can link bank, card, loan, and investment accounts through Plaid; Claude's access is read-only (balances, credit limits, and up to 24 months of transactions) and cannot move, manage, or invest money. Chats in the Personal Finance hub are not used for model training. [Confirmed; source: Anthropic] https://claude.com/resources/articles/introducing-personal-finance-in-claudehttps://ohmyai.ai/archive#anthropic-claude-enterprise-security-inference-hooks-ga-2026-11-05https://ohmyai.ai/archive#anthropic-claude-enterprise-security-inference-hooks-ga-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedSafety & securityProductsAnthropic said Claude Enterprise now includes a security center (in beta) that compares an organization's Claude settings with Anthropic's recommended configuration, plus generally available inference hooks. Prompt hooks and pre-tool-call hooks send prompts and tool calls to an organization's security tools for an allow-or-deny check before generation or tool execution. The update also covers centralized management and additional access controls for Enterprise admins. [Confirmed; source: Anthropic] https://claude.com/resources/articles/claude-enterprise-security-updateshttps://ohmyai.ai/archive#meta-muse-realtime-voice-avatars-us-2026-11-05https://ohmyai.ai/archive#meta-muse-realtime-voice-avatars-us-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedProductsModels & launchesMeta started rolling out Muse Realtime Voice and Muse Realtime Avatar in the Muse app on mobile, web, and Mac in the United States. Users can start a call from the phone icon, interrupt Muse, choose voice-only or avatar display modes, and pick from hundreds of voices; Meta says voice on AI Glasses and Muse Charm is coming soon. [Confirmed; source: Meta] https://about.fb.com/news/2026/11/introducing-muse-ai-voice-expressive-avatars/https://ohmyai.ai/archive#thinking-machines-inkling-2-small-2026-11-05https://ohmyai.ai/archive#thinking-machines-inkling-2-small-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedModels & launchesOpen sourceThinking Machines Lab launched Inkling 2 Small, a multimodal Mixture-of-Experts model with 594 billion total parameters and about 17 billion active, with image input and variable effort. It is available now through Tinker with closed weights; the company says full Apache 2.0 weights are planned after external safety testing and red-teaming. [Confirmed; source: Thinking Machines Lab] https://thinkingmachines.ai/news/inkling-2-small/https://ohmyai.ai/archive#mistral-studio-managed-agents-apps-2026-11-05https://ohmyai.ai/archive#mistral-studio-managed-agents-apps-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedAgentsProductsMistral added managed agents and apps to Mistral Studio in public preview, alongside a natural-language workflow builder, HTTP API connectors, tracing, evaluations, and related Studio tooling. Managed agents can run for long periods with durable sessions and human approvals; apps can be scaffolded and deployed with a CLI; Studio can run in cloud, VPC, or on-premises. [Confirmed; source: Mistral] https://mistral.ai/news/build-with-studio/https://ohmyai.ai/archive#perplexity-teammates-always-on-agents-2026-11-05https://ohmyai.ai/archive#perplexity-teammates-always-on-agents-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedAgentsProductsPerplexity introduced Teammates, always-on agents that can take multiple ongoing roles, use Perplexity Computer projects, skills, and memory, and collaborate in Slack, with progress tracked in per-agent threads. The feature requires Perplexity's Max plan. [Confirmed; source: The Verge] https://www.theverge.com/ai-artificial-intelligence/1028302/perplexity-teammates-ai-agents-workhttps://ohmyai.ai/archive#cursor-build-live-apps-cloud-agents-2026-11-05https://ohmyai.ai/archive#cursor-build-live-apps-cloud-agents-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedAgentsProductsCursor's Build Live Apps release lets a cloud agent scaffold a web app, provision Postgres if needed, set up auth behind Cursor SSO, and deploy to a shareable URL after security checks pass and the user explicitly publishes. It is available on paid plans with cloud agents enabled; Origin and GitHub repos are supported, with GitLab support planned. [Confirmed; source: Cursor] https://cursor.com/changelog/build-live-appshttps://ohmyai.ai/archive#lithuania-seimas-ai-supervision-law-2026-11-05https://ohmyai.ai/archive#lithuania-seimas-ai-supervision-law-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedPolicy & regulationOn November 5, 2026, Lithuania's Ministry of Economy and Innovation said the Seimas had adopted a law on the supervision of artificial intelligence that puts the EU AI Act into practice nationally. The ministry will coordinate AI policy, the Innovation Agency will run an AI regulatory sandbox, and the State Digital Solutions Agency will give methodological help to public-sector bodies. Existing sector market-surveillance authorities will oversee high-risk AI systems in their fields, and the Communications Regulatory Authority (RRT) will be the single point of contact. The ministry said the law takes effect on December 1, 2026. [Confirmed; source: Lithuanian Ministry of Economy and Innovation] https://eimin.lrv.lt/lt/ziniasklaidai/naujienos/seimas-prieme-dirbtinio-intelekto-prieziuros-istatyma-tWA0/https://ohmyai.ai/archive#eu-education-skills-package-ai-2026-11-05https://ohmyai.ai/archive#eu-education-skills-package-ai-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedPolicy & regulationOn November 5, 2026, the European Commission launched an Education and Skills Package aimed at declining basic skills and rapid digital change. It sets four priorities: basic skills, digital and AI tools for education, the teaching profession, and vocational education and training. The Commission noted that a European Digital Infrastructure Consortium for education (EDU-EDIC) was set up on November 3 at the request of eight member states to develop trusted AI and digital services for schools. [Confirmed; source: European Commission] https://digital-strategy.ec.europa.eu/en/news/commission-launches-education-and-skills-package-tackle-declining-basic-skills-rapid-digital-changehttps://ohmyai.ai/archive#box-nvidia-nemotron-context-retrieval-architect-2026-11-05https://ohmyai.ai/archive#box-nvidia-nemotron-context-retrieval-architect-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedBusiness & investmentAgentsOn November 5, 2026, Box announced Box Context Retrieval, a content-retrieval system for AI agents built on Nvidia's Nemotron models, which it plans to make generally available to Enterprise Advanced customers early next year. Box says that in its testing the system improved search accuracy by about 20%, used about 90% fewer tokens and answered questions about four times faster. Box also unveiled Box Architect, a tool for building AI-driven content workflows. [Confirmed; source: Reuters (via US News)] https://money.usnews.com/investing/news/articles/2026-11-05/box-expands-nvidia-partnership-unveils-ai-workflow-builderhttps://ohmyai.ai/archive#pfn-plamo-3-translate-open-weights-2026-11-05https://ohmyai.ai/archive#pfn-plamo-3-translate-open-weights-2026-11-05Thu, 05 Nov 2026 12:00:00 +0000ConfirmedOpen sourceModels & launchesJapan's Preferred Networks released the weights of its PLaMo 3 Translate machine-translation models in 0.8B, 2B, 8B and 31B parameter sizes, supporting 53 languages. The models are published on Hugging Face under the PLaMo Community License, which allows free non-commercial use and free commercial use for organizations with annual revenue of up to 1 billion yen after registration. [Confirmed; source: Preferred Networks] https://www.preferred.jp/en/blog/tech/plamo-3-translate-weight-release