Twenty-Nine Countries Just Built Their Own AI Governance Body. The US and Western Europe Are Not in It.

Work smarter, not longer.

Why spend time figuring out the perfect prompt when experts have already done the work for you?

Get 100 ready-to-use ChatGPT prompts for writing, brainstorming, research, emails, marketing, and productivity—designed to help you get more done in less time.

1  Moonshot AI released Kimi K3, which immediately claimed the top spot on Arena's web design leaderboard, above Fable 5 and GPT-5.6.

Kimi K3 is Moonshot AI's most capable model yet, built for coding, knowledge work, and general reasoning at competitive cost. Within hours of launch it topped Arena's web design code leaderboard, overtaking both Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol. Moonshot is reportedly in talks to raise at a $31.5B valuation, which would make it one of the most valuable AI labs outside the US. Watch the launch video (1M views) or try Kimi K3 here.

2  Twenty-nine countries signed an agreement to establish a new intergovernmental AI body headquartered in Shanghai.

The World AI Cooperation Organization counts China, Russia, Brazil, Venezuela, Cuba, Belarus, Serbia, ten African nations, and twelve Asian nations among its founding members. The United States and all Western European countries are absent. A global AI governance body that does not include the US or Europe is not a governance body for the global AI ecosystem. It is the beginning of a parallel one.

3  Canva added vibe-coding to its platform: describe what you want to build, and it generates working code you can edit like any Canva design.

Canva Code 2.0 lets anyone build websites and apps from a plain-language prompt, then edit the output using Canva's familiar design interface. You can start from a template, paste in existing HTML, or carry over another Canva project. It is available now on all plans, including free. See how it works. The gap between designing something and shipping something just got significantly narrower for anyone already in the Canva ecosystem.

Legal. Apple filed a federal lawsuit against OpenAI alleging a coordinated campaign to poach its engineers and steal proprietary hardware secrets. The suit names two former Apple employees now at OpenAI's hardware division, accusing the company of coaching them to funnel confidential knowledge across. Apple called the hardware division "rotten to its core." OpenAI denies everything. The case lands as OpenAI is preparing to go public, which makes the legal exposure considerably more than a distraction.

Governance. Google DeepMind CEO Demis Hassabis published a rare essay calling for a FINRA-style self-regulatory body to track frontier AI progress and review new models before public release. The proposal is notable for where it comes from: one of the most credible voices in the field arguing that the industry should regulate itself rather than waiting for government to impose structure. His timeline is the end of this year.

Open vs. closed. Mira Murati's Thinking Machines shipped Inkling, the most competitive open-weight model yet to come from a US lab. It arrives as American enterprises push to own more of their AI stack rather than depending on proprietary APIs. Combined with Kimi K3 debuting at the top of Arena's coding leaderboard, this week saw the open-weight category become genuinely competitive with closed frontier models for the first time.

Demand vs. efficiency. Former Intel CEO Pat Gelsinger told CNBC that AI demand remains "almost unlimited," with Nebius CRO Marc Boroditsky echoing that supply cannot keep up. The nuance both flagged is worth holding alongside that headline: companies are getting considerably more deliberate about which AI investments they fund. Unlimited demand at the category level does not mean unlimited budgets at the project level. Those are two different things, and the distinction is now showing up in how enterprise software stocks are trading.

Thin prompts, fat context. Anthropic engineer Thariq Shihipar shared his ideal prompting strategy: keep the prompt itself short, but load it with rich context — examples, background, constraints, and specifics about what you want. The thread sparked a sharp discussion in the replies about when this works and when it doesn't.

Flight simulator, Claude edition. A Redditor built a fighter jet game entirely with Claude Code that one commenter described as "the most ridiculous and insane thing I've seen all day." It rivals Microsoft Flight Simulator. The build process is documented in the thread (2K upvotes).

Kimi K3 vs. Opus 4.8 head-to-head. An AI power user ran both models through a one-prompt video game scene challenge and shared the outputs side by side. The comment section agrees Kimi K3 won, and the margin surprised people who were not expecting a Chinese lab's newest model to pull that clearly ahead of Opus 4.8 on a creative task.

The Haaland deepfake. A video of Erling Haaland appearing to startle at his own reflection went viral at 13M views. A follow-up post showed the side-by-side comparison suggesting the clip was AI-generated. The original still looks convincing at normal playback speed. It is a useful benchmark for where consumer-grade AI video sits right now.

Claude solves a six-month physics problem. Yuji Tachikawa, one of the world's leading theoretical physicists, posted that Claude Fable solved a research problem he had been stuck on for six months. The Reddit thread has 2,700 upvotes and a comment section that ranges from impressed to sceptical in roughly equal measure.

Turn a folder of PDFs into a single structured summary document

This Claude Cowork prompt processes an entire folder of PDFs in one session, producing a structured summary for each file with four consistent sections: Purpose, Key Findings or Terms, Action Items or Red Flags, and a Confidence Rating on how complete the document appears. It then compiles all summaries into a single Word document saved to a folder you specify. Designed for research stacks, legal review, or any situation where you need to move through a large number of documents quickly without reading every page.

Replace [FOLDER PATH] with the exact location of your PDFs on your computer. Replace the document type description with what you are actually working with, since naming the document type helps the model calibrate what to look for. Replace [OUTPUT FOLDER PATH] with where you want the compiled Word file saved. For legal documents, add one instruction after the main prompt: "Flag any clause that restricts assignment, limits liability, or contains a termination trigger." That single addition captures the Red Flags that matter most in a contract review without requiring the model to guess at your priorities.

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide