Hello AI Enthusiast,
This week, Anthropic revealed a model so capable at finding security vulnerabilities that it decided not to release it. Microsoft quietly announced it no longer needs OpenAI to build competitive AI. And OpenAI published a policy paper proposing robot taxes and four-day workweeks. A busy week for the industry's biggest names.
The Big Picture 🔊
Anthropic's New Model Is Too Dangerous to Release
Anthropic launched Project Glasswing, giving a restricted group of 40 organizations, including Apple, Google, Microsoft, and Amazon, access to Claude Mythos Preview, its most capable and unreleased model. The AI has already found thousands of critical vulnerabilities across major operating systems and browsers, some decades old. Anthropic is not releasing Mythos publicly, citing the risk that such capabilities could be weaponized.
|
||
|
If the numbers are real, this is not marketing. A model that finds hundreds of zero-day vulnerabilities in critical infrastructure is genuinely dangerous in the wrong hands, and withholding it until defenders can act makes sense. Though the framing is worth noting: Anthropic is saying this model creates a global security problem, and the only solution is also this model. |
Microsoft Builds Its Own AI Models
Microsoft released three in-house AI models: MAI-Transcribe-1 for speech-to-text, MAI-Voice-1 for voice generation, and MAI-Image-2 for image creation. Built by Mustafa Suleyman's MAI Superintelligence team, the models are available on Microsoft Foundry and priced to undercut competitors. This is a sign of Microsoft reducing its dependence on OpenAI, six months after renegotiating a contract that previously blocked it from building its own models.
|
||
|
Mustafa Suleiman built Pi, one of the most thoughtful AI products of 2023. Microsoft acquired his company just to hire him, then shut it down. The new microsoft.ai branding feels like a signal. With the OpenAI contract out of the way, we might finally start seeing his vision for user-friendly AI reflected in Microsoft products. |
The biggest AI companies are racing to build more powerful models. Most teams are still figuring out the ones they already have. Our Corporate AI Training helps bridge that gap.
OpenAI's Blueprint for the AI Economy
OpenAI released a 13-page policy paper proposing 20 ideas for managing AI's economic impact, including public wealth funds, robot taxes, expanded safety nets, and a four-day workweek pilot. The document frames itself as a starting point for debate, not a final blueprint. Critics point out that most ideas are familiar, and some question the timing, given a major investigative piece about Sam Altman published the same day.
|
||
|
These proposals sound reasonable until you try to argue the opposite. Who would publicly say prosperity should not be shared, or that risks should not be managed? When you cannot disagree with something, it is usually not saying much. Governments are behind on AI policy, that part is true. |
Bits and Bobs 🗞️
Google now offers free video creation with its AI-powered Veo 3.1 model, allowing users to generate high-quality videos from simple prompts or photos.
Gemma 4 is an open-source AI model family optimized for efficient, on-device and local hardware use, enabling developers to fine-tune and deploy powerful multimodal AI with low latency and full control.
Cursor 3 introduces a unified AI-driven workspace with seamless cloud and local agent collaboration to simplify and enhance software development.
French AI startup Mistral secured $830 million to build a large data center with 13,800 Nvidia GPUs to train and operate its AI models in Europe.
Google's Gemini AI now features a redesigned, expert-developed interface that connects users showing signs of mental health crises directly to crisis hotlines and professional support.
Anthropic is expanding its partnership with Google and Broadcom to secure multiple gigawatts of next-generation TPU capacity starting in 2027 to power its Claude AI models and meet growing global demand.
OpenAI's new voice-first ChatGPT for Apple CarPlay enables hands-free AI interaction while driving but cannot access maps, vehicle data, or control other apps.
From Our Founder’s Channels 🤳
In his latest LinkedIn post on artists starting the AI resistance, Gianluca Mauro argues that while some companies will profit from AI-automated convenience, others will profit from human inefficiency and absolute dedication to craft. Read it here.
LOLgorithms 😂
Windows XP for the win!
That's a wrap on our newsletter! Before you go, here’s a quick recap of our offerings:
AI Academy Membership: Get 12 months of access to all our cohort-based programs and on-demand courses.
AI Agent Bootcamp: Accelerate processes and solve business problems by building AI Agents, without coding.
Practical Introduction to ChatGPT: A free course on using ChatGPT confidently, understanding its workings, and exploring its potential.
Customized Corporate Training: Equip your team with the skills they need to unlock the potential of AI in your business.
Catch you next week! 👋
