Good morning, this is Daily AI News for Saturday, November 21st. Five stories in about three minutes. First, Gemini can now use your Mac. Google's assistant can now work in the background with your files, apps and websites. That makes it an AI agent, which simply means an AI that takes actions, not just answers questions. For example, you could ask it to find last month's receipts in your downloads, add up the totals in a spreadsheet, and save it to a folder. For now it's English only, in the US, for paid AI Plus, Pro and Ultra plans. Why it matters: AI is moving out of the chat window and into the apps you already use. Next, a small model with big skills. Thinking Machines released Inkling 2 Mini with open weights. Open weights means the trained model itself is free to download and run on your own computer. It has 26 billion parameters, the internal settings a model learns during training, but only about 3.4 billion are active at once, so it's light enough for a laptop. Among open models under 100 billion parameters, it leads coding tests. Picture a student fixing code on a train with no internet, and nothing leaves the laptop. Why it matters: strong AI that runs privately and offline is becoming normal. Third, Microsoft launched seven of its own AI models, called MAI. They cover coding, security, images, live speech and transcription. Microsoft says its coding models cost up to 95 percent less than rivals, and they're already inside GitHub Copilot, a popular coding assistant. Why it matters: when AI gets cheaper to run, the apps you use can add more AI features without raising prices. Fourth, Europe set clearer rules for AI labels. The EU published guidelines for Article 50 of its AI Act. They explain when you must be told you're talking to a chatbot, and when AI-made content and deepfakes must be labelled. A deepfake is a fake but realistic video, image or voice made with AI. Labels must also be machine-readable, like a hidden tag that software can detect. Imagine scrolling social media and seeing a clear note that a politician's video was made by AI. Why it matters: it should get easier to tell what's real online. And finally, OpenAI is cracking down on distillation. Distillation is when someone trains their own model on the answers of a stronger model, like a student copying the teacher's answer sheet. Meanwhile, a US congressional report calls for Chinese open-weight models to be removed from federal systems and critical infrastructure. Why it matters: top AI models are becoming strategic assets, and access to them is tightening for companies and governments. That's your AI briefing for today. Sources are in the show notes. See you tomorrow. This episode was read by an AI voice.