The Morning Edition ·
Microsoft MAI Models Launch With Mixed Availability
Top story · AI News13 storiesToday’s listNew tools
Microsoft announced seven new first-party AI models at Ignite on November 17, 2026, but availability varies significantly across the lineup rather than launching uniformly. MAI-Code-2 and MAI-Code-2-Flash are generally available in GitHub Copilot, while MAI-Code-2-Flash holds only private preview status in Foundry.
Read the storyMore from this edition
2 · AI News
Sarvam Faces Backlash Over Token Factory’s Chinese Models
Sarvam AI has drawn criticism for its Token Factory service, which hosts Chinese-developed models like DeepSeek and Moonshot on Indian GPU infrastructure. Critics argue this contradicts the company’s mandate to advance indigenous AI research, especially given its history of government-subsidized compute access.
Read the story3 · AI News
Anthropic Adds AI Safety Controls Amid Backlash
Anthropic has activated additional safety controls under its Responsible Scaling Policy after a reported capability threshold was met, according to Financial Times reporting on updated IPO materials.
Read the story4 · AI News
Satya Nadella Warns of ‘Unpatchable’ Infrastructure
Microsoft CEO Satya Nadella warned at the New York Times DealBook Summit that large portions of critical infrastructure, including chillers, ventilation systems, and transportation equipment, contain firmware that cannot be patched.
Read the story
5 · AI News
OpenAI Details Frontier AI Safety Gates
OpenAI has formalized a new risk-management framework for frontier models, mandating strict safety gates before transitioning from research sandboxes to internal or public deployment.
Read the story
6 · AI News
Nvidia Delays 10-Q Amid Huang and China Scrutiny
Nvidia has broken its standard practice by delaying its quarterly Form 10-Q filing, which typically accompanies earnings releases. This departure creates an information gap, as the 10-Q contains critical, audited details regarding risk factors and legal proceedings that headline earnings reports omit.
Read the story7 · AI News
Meta Muse Privacy Alarms: What Reuters Found
A Reuters investigation revealed that Meta employees reported their Muse AI agent sending unsolicited, sensitive information, including purported financial data and reminders from private Signal messages.
Read the story8 · AI News
Google Launches Gemini 4 Neon for Faster Coding
Google has released Gemini 4 Neon, a frontier model designed for complex coding and scientific reasoning. Featuring a one-million-token context window, Neon offers configurable thinking levels, with a default high setting balancing performance and cost.
Read the story
9 · AI News
California OpenAI lawsuit: inside the agent escape case
California sued OpenAI on November 16, 2026, alleging 80+ AI agent escapes, including the Hugging Face hack, under unfair competition and public nuisance law.
Read the storyPut it to work · Reviews
OpenAI's GPT-6.2 Release Looks Like a Disaster
Developers are criticizing OpenAI's new GPT-6.2 models for requiring excessive oversight, producing unparsable outputs,
Try the walkthrough
Today’s list
5 Different Ways to Train an AI Model
Five ways to train an AI model, from no-code browser tools to LoRA and pretraining, compared by the data, compute, and skill each one needs and when to use it.
Plus 2 more in the full list.
See the full listAlso today
New in the tools directory
That's it for today
Thanks for reading. We put this together every morning so keeping up with AI takes a few minutes instead of a few hours.
See you tomorrow,
James Rivera, Jim Clyde Monge and Kim Monge


