Here are today's news for 22nd of October about Meta, Anthropic, OpenAI and Google Cloud. Let's dive into the details. On October 21, 2026, Meta said it would create publicly traded, non, voting Class C shares to give it more flexibility to fund AI expansion, employee compensation and acquisitions without weakening existing shareholders’ voting power. Effective January 19, 2027, shareholders are due one Class C share for each Class A or B share, a distribution expected to work like a two, for, one stock split. The dividend itself raises no cash; Meta can raise capital by issuing additional Class C shares later. The announcement came with third, quarter results and a higher 2026 capital, expenditure forecast: $135 billion to $145 billion, with the lower bound raised from $130 billion. The scale of that investment reflects fierce competition for AI infrastructure and talent, while raising the stakes for investors if spending does not generate returns. CEO Mark Zuckerberg said more than 10 million people in the U.S. and Canada had set up Muse personal agents in about six weeks, and called its growth strong. On October 21, Anthropic opened a waitlisted US research preview of Claude Duet, a proactive work agent for Claude Pro and Max subscribers. Available on iOS, Android and claude.ai, Duet sends a morning brief, monitors connected apps and can handle work from scheduling meetings to building dashboards, forecasts and web apps. It is built around Claude Opus 5.5, uses a virtual machine with a browser to keep working in the background, and draws on Claude account memory, skills and more than 6,000 Marketplace connectors. Duet shares subscribers’ existing usage limits; background checks count, while larger projects consume more model turns. Anthropic says it asks approval before sensitive actions, but warns it may misread messages or misjudge urgency. The launch puts Anthropic alongside OpenAI’s Dots and Meta’s Muse in a race to make agents that act rather than wait for prompts. On Oct. 21, 2026, OpenAI and Google Cloud announced that three OpenAI models, GPT, 6 Astra, GPT, 6.1 Sol and GPT, 6 Luna, were entering private preview in Google Cloud’s Model Garden for selected enterprise customers. Through Google Cloud APIs, customers can also connect local Codex CLI and Desktop workflows to Google’s platform for inference. General availability is planned “in the coming weeks.” The deal lets businesses access OpenAI models through Google Cloud’s governance and infrastructure, buy them via Google Cloud Marketplace and apply existing GCP spending commitments; usage appears on one cloud bill. That matters particularly to regulated enterprises and organizations already committed to Google Cloud, while intensifying competition among cloud providers to distribute leading AI models. Google describes Astra as built for demanding coding, computer, use, professional and scientific work; Sol as near, Astra intelligence at lower cost; and Luna as a fast, cost, efficient option for everyday workloads. On 8 October 2026, the UK Information Commissioner’s Office said ten foundation, model developers, Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI, had made, or committed to make, data, protection changes after two years of regulatory scrutiny. The changes cover clearer explanations of personal, data use in model development, stronger ways for people to exercise their rights, and more rigorous assessments of safeguards; the ICO says it is monitoring delivery, not declaring every issue resolved. Its supervision found gaps in how companies explained purposes and assessed whether safeguards actually reduced risks to people. Some firms have published or updated training, data privacy notices, including information about data sources, development stages, retention and international transfers. Microsoft Research announced on October 21, 2026, that senior researcher Xinzhi Zhang and collaborator Zhao Song had proved the 1995 Kannan, Lovász, Simonovits (KLS) conjecture. Their paper, An O(1) Bound for the KLS Constant, posted in final form on October 4, shows that the KLS constant, and the Poincaré constant for isotropic log, concave probability measures, has a universal bound independent of dimension. The result says simple flat cuts capture the narrowest bottlenecks of these distributions within a fixed factor, strengthening foundations for high, dimensional sampling, optimization and statistics. The work could eventually improve algorithms used across machine learning and scientific computing, though it is a theoretical advance, not a product launch or immediate performance claim. Zhang says the team’s human, directed AI, agent search explored more than 100 candidate approaches; researchers pruned ideas, developed the recursive argument and verified the proof, while AI assisted with derivations, drafting and internal checks. That's all for today's AI news briefing. Have a great day!