Trump’s ‘ONE RULE’ Collides With China Chips

Welcome back! Today’s AI stories are all about who’s grabbing the steering wheel. Presidents are trying to write a single rule that bends fifty different power centers, lab bosses are yanking roadmaps around one scary benchmark, researchers are turning nursery-rhyme jailbreaks into stress tests and designers are quietly sketching with new canvases while everyone else argues about risk. The models change names every quarter but the real plot today is how deals, exploits and doodles are starting to shape what AI actually becomes.

In today’s Generative AI Newsletter:

On December 8, Donald Trump pulled two AI levers at once. He pledged an executive order to impose a single national 'one rule' for AI, arguing that chasing 50 approvals from different states would limit innovation. At the same time, he cleared NVIDIA to sell its H200 chips to approved buyers in China with a condition that Washington takes a 25% cut of every sale. One president and a few giant firms now sit at the junction of state rights, chip exports and everyday AI risks.

How the regulation looks as a test of AI rule-by-deal:

The one rulebook order and the H200 bargain sketch an AI doctrine built around speed, scale and White House deal-making. Simpler rules and a reopened export channel could encourage more companies to comply with the national rulebook, leading to greater standardization and oversight in the AI industry. However, the loudest warnings from statehouses, Congress and security officials come from outside the room where the terms are cut.

_{Special highlight from our network}

What if you could run thousands of tools straight from ChatGPT, without writing code or wiring integrations?

Now you can.

With ChatGPT Atlas + Apify MCP, one connector gives you access to over 10,000 workflows:
Scrape data. Run APIs. Monitor anything. Pipe results to Sheets or dashboards.
No dev time. No complex setups. Just answers, fast.

It’s built for research, enrichment, and automation at scale. And it works inside the tools your team already uses.

What would you automate first?  

_{Special highlight from our network}

You know the feeling — the meeting ends, and suddenly you’ve got homework. Notes to tidy, action items to assign, emails to draft. Radiant handles all that for you.

This free AI meeting assistant quietly captures your meetings (without a bot in sight) and drafts your follow-up emails for you. Summaries, next steps, and action items land in one neat draft you can open in Gmail, review, and send in seconds.

Meetings in, momentum out. We like that equation.

Download for Mac

Sam Altman recently told the staff that ChatGPT was in code red after weeks of complaints about slower answers. Google’s Gemini 3 had topped 'Humanity's Last Exam' a benchmark for broad reasoning, so a GPT 5.2 release planned for later in the month was pulled forward to as early as next week. OpenAI says this upgrade will fix slow answers while it also signs a multi-billion-dollar data center plan in Australia to host a huge GPU cluster.

Here is how that rush looks inside OpenAI right now:

Google once declared a code red because of ChatGPT but now OpenAI is returning the favor by rushing out GPT-5.2. It has wired itself into higher global costs and irritated subscribers. If GPT 5.2 gives faster, clearer help, others will copy the crisis playbook. If it lands like another patch, "code red" may sound less like safety and more like how AI gets shipped. The success or failure of GPT-5.2 will likely set a precedent for future AI developments across industries.

A study from Italy’s Icaro Lab shows that simple poems can coax advanced models into unsafe territory. The researchers turned harmful prompts into verse and discovered that models treated the poetic structure as a kind of disguise. The technique worked across labs and model families, revealing a blind spot where creativity behaves like an exploit. The result feels both amusing and unsettling, and the data highlights how easily style becomes a pathway around safety tools.

Here Is What the Study Found:

The researchers declined to share the poems, describing them as too dangerous and surprisingly simple to construct. The study suggests that models struggle when harmful instructions arrive wrapped in unfamiliar linguistic patterns. It also adds poetry to a growing list of creative jailbreaks that slip past guardrails through structure rather than intent. The team described the work as evidence that safety remains an evolving puzzle. Every improvement reshapes the threat landscape, and this round belongs to the poets who turned riddles into a stress test for modern AI.

_{Special highlight from our network}

Retailers are racing to adopt agentic AI, but trust and human experience still form the heart of a great retail experience. Faster journeys do not change what shoppers want. They want clarity and control. 

The best teams look for moments where an agent can remove friction without stripping dignity. That might mean clearer comparisons, surfacing a return policy early or giving staff context so conversations feel natural. 

Thoughtworks has opened new research on where these systems work well and where they fail. If you want a practical view beyond hype, the session is worth your time. Which part of retail is most ready for agents, and which should stay human?

Sign-up here

Mixboard is an experimental Google Labs tool that acts like an AI powered moodboard. You start with a blank canvas or a prompt, then mix text, images, doodles, and AI generated visuals to explore early ideas. It works well for quick concepting, design experiments, event planning, DIY projects, or anything that benefits from visual brainstorming.

Core functions (and how to use them):

Try this yourself:

Start a board for a real project. Upload one or two reference images, add a few AI generated ideas, then doodle a change on one of them. Ask Mixboard to turn the board into a presentation and review how your scattered thoughts are now structured into something clearer.