Daily AI News

AI News Today: Gemini on Mac, Microsoft MAI Models, EU AI Act (Nov 21, 2026)

Daily briefing · Saturday, November 21, 2026 · 3:12

AI news today: Google Gemini, Thinking Machines, Microsoft MAI and the EU AI Act, in about 3 minutes.

Gemini can now use your Mac, Microsoft launches seven MAI models in GitHub Copilot, and the EU issues Article 50 AI Act transparency rules, plus OpenAI, ChatGPT Live Dictation and more, in today's Daily AI News for Saturday, November 21, 2026.

In this episode

1. Gemini can now use your Mac

Why it matters: Desktop AI agents are starting to work in the apps and files people already use.

Source: TechCrunch

2. Thinking Machines has released Inkling 2 Mini

Why it matters: A strong open model that runs on a laptop makes local coding help more widely available.

Source: Thinking Machines

3. Microsoft has launched seven MAI models

Why it matters: Cheaper in-house models in GitHub Copilot could cut what teams pay for everyday AI coding help.

Source: Interesting Engineering

4. The EU has published guidelines for Article 50 of the AI Act

Why it matters: Anyone using chatbots or synthetic media in the EU now has clearer rules on when they must disclose it.

Source: Il Sole 24 Ore

5. An AI-assisted cyberattack hit the European Commission

Why it matters: AI-assisted attacks are already being used against major public institutions, not just in theory.

Source: AI.info

6. OpenAI is cracking down on distillation

Why it matters: Access to leading models, and to Chinese open weights, is tightening for companies and governments.

Source: OpenAI, Homeland Security Committee

7. ChatGPT has gained Live Dictation

Why it matters: Voice input is getting easier in ChatGPT, while some creative tools now let people switch AI off.

Source: The Neuron

Chapters

  1. 0:00 Intro
  2. 0:13 Gemini can now use your Mac
  3. 0:35 Thinking Machines has released Inkling 2 Mini
  4. 1:03 Microsoft has launched seven MAI models
  5. 1:29 The EU has published guidelines for Article 50 of the AI Act
  6. 1:57 An AI-assisted cyberattack hit the European Commission
  7. 2:18 OpenAI is cracking down on distillation
  8. 2:42 ChatGPT has gained Live Dictation
  9. 3:05 Outro

Full transcript

This is Daily AI News, Saturday, November 21st. Gemini takes over your Mac, Microsoft drops seven new models, and Europe sets the rules for AI labels. Let's go.

First up. Gemini can now use your Mac. Google's agent works in the background with files, apps and websites. For now it is English-only, in the US, for AI Plus, Pro and Ultra subscribers. Desktop AI agents are starting to work in the apps and files people already use.

Next. Thinking Machines has released Inkling 2 Mini with open weights. The model has 26 billion parameters, of which 3.4 billion are active, and it is small enough to run on a laptop. Among open models under 100 billion parameters, it leads coding benchmarks such as DeepSWE. A strong open model that runs on a laptop makes local coding help more widely available.

Also today. Microsoft has launched seven MAI models. They cover coding, security, images, real-time speech and transcription, and according to Microsoft they cost up to 95 percent less than competitors for coding. The coding models are already available in GitHub Copilot. Cheaper in-house models in GitHub Copilot could cut what teams pay for everyday AI coding help.

Meanwhile. The EU has published guidelines for Article 50 of the AI Act, which is about transparency. They explain when people must be told they are talking to an AI, and when AI-generated content and deepfakes must be labelled, including in machine-readable form. Anyone using chatbots or synthetic media in the EU now has clearer rules on when they must disclose it.

And this. An AI-assisted cyberattack hit the European Commission. Several systems were taken down for a while, including payroll and reimbursement of health expenses. Most are back up, but the investigation continues. AI-assisted attacks are already being used against major public institutions, not just in theory.

Plus. OpenAI is cracking down on distillation, where others train their own models on the outputs of OpenAI's models. Meanwhile, a US congressional report calls for Chinese open-weight models to be removed from federal systems and critical infrastructure. Access to leading models, and to Chinese open weights, is tightening for companies and governments.

And finally. ChatGPT has gained Live Dictation, which shows the text as you speak and no longer has a ten-minute limit. Canva has also added a switch that turns off most of its AI tools. Voice input is getting easier in ChatGPT, while some creative tools now let people switch AI off.

That's your AI news for today. Sources in the show notes. See you tomorrow.

Sources

About Daily AI News: a daily AI news briefing on the most important news in artificial intelligence, in about three minutes. The voice in this episode is synthetic and AI-generated, and the script was prepared with the help of AI. Plain-text transcript.

All episodes · RSS feed