Matt Wolfe · Video summary

AI News: All AI Labs Want To Slow Down (Except One)

This week in AI, the major AI labs respond to concerns about AI's potential risks, with Anthropic's CEO proposing a three-part plan for safe development. Other companies like Meta, Google, and Apple also announce new AI features and products.

Matt Wolfe · September 18, 2026 · 27-min episode · 4-min read

AI News: All AI Labs Want To Slow Down (Except One)

Key takeaways

📜Anthropic's CEO, Dario Amodei, proposes a three-part plan for responsible AI development, including embedded evaluators, regulation within democracies, and global cooperation.00:40
🤝Major AI leaders like Sam Altman and Satya Nadella express agreement with Anthropic's approach to pacing AI development and implementing independent evaluators.03:48
💡Mark Zuckerberg of Meta takes a different stance, advocating for labs to self-regulate and prioritize safety without external mandates, citing Meta's own practices.04:46
🔄Anthropic is merging Claude Cowork and Chat into a single interface and redesigning Projects for a more conversational workflow, though these features are rolling out gradually.09:37
🗣️Google's Gemini Notebook will soon allow real-time conversations with notes and audio recording of lectures, offering a powerful tool for students.13:13
🍎Apple announces a more capable Siri AI powered by Apple Intelligence, with features like screen context awareness and email drafting, but it's also rolling out in beta.14:19

Chapters

0:17AI Labs Respond to Risk Concerns
0:40Anthropic's "We Must Pace the Frontier" Article
5:40Industry Reactions to Anthropic's Plan
15:37Claude Cowork and Chat Merge
17:33Claude Projects Redesigned
21:45Gemini Notebook Updates
23:39Apple's Siri AI Update
32:21Other AI News and Updates

The digest

Last week, the AI world was dominated by discussions from Anthropic and OpenAI employees expressing concerns about the potential risks of advanced AI, including the possibility of it destroying humanity. This week, the narrative shifted as major AI labs began to respond to these concerns, outlining their strategies for responsible AI development.

Anthropic's Three-Part Plan for Pacing the Frontier

Dario Amodei, CEO of Anthropic, published an article titled "We Must Pace the Frontier," which details a comprehensive three-part plan for managing AI development safely. Amodei acknowledges the risks associated with AI and states that Anthropic has prioritized caution over speed and profit. His plan includes:

Embedded Evaluators: Anthropic commits to inviting third-party evaluators who will have access similar to internal employees, including desks, access badges, and company laptops, to conduct comparable risk assessments. These evaluators will operate independently to ensure unbiased safety reviews.
Pacing Within Democracies: Amodei proposes regulations targeting all US frontier AI companies, including those unwilling to cooperate voluntarily. He emphasizes that these regulations must also ensure the US maintains its lead over countries like China, specifically by restricting the sale of powerful AI chips and manufacturing equipment to China, cracking down on unauthorized distillation of models, and strengthening security against model theft.
Global Pacing: The final part of the plan calls for international cooperation, particularly with China, to establish global agreements on AI development. This includes prohibiting dangerous uses of AI (like biological weapons), agreeing to test models before release, and implementing a "speed limit" on the rate of recursive self-improvement in AI models.