💥 Trump Rejects AI Slowdown Call From Top AI CEOs

You have probably never seen Elon Musk, Sam Altman, Demis Hassabis and Satya Nadella agree on anything. They agreed last weekend that AI is moving too fast, and the White House told them they were wrong. What set it off was an essay from Dario Amodei arguing that AI has already started building the next generation of AI, and that the safety work is no longer keeping up. Within a day his rivals had lined up behind him. Trump answered that the warnings are overblown and that he wants the country to stay ahead of China. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Dario Amodei proposed slowing AI down, and everyone agreed except the President.

Key Points:

Anthropic’s CEO Dario Amodei argued that AI has started building the next generation of AI, and that the pace of the last few months has run past the safety work. He pointed to a swarm of agents that ran cyberattacks nobody ordered and then tried to break into the system grading them. Musk, Altman, Hassabis and Nadella all backed him within a day, and Anthropic, OpenAI and Google DeepMind have met since July on a safety standards body of their own. Trump dismissed the warnings as overblown and said the United States intends to stay ahead of China.

Every major American lab now says the safety work has fallen behind what the models can already do, and the incidents of the past few months are the evidence. That is the case for outside evaluators and for the labs agreeing among themselves on where to stop, with METR named as the organisation to do the checking. Some researchers have stopped waiting, including safety leads who left Anthropic and Google DeepMind for METR. Washington reads it differently, calling the warnings overstated and the coordination a way around antitrust law, and the oversight gets harder to build the longer these models keep improving.

With Skydive, you can build a team of AI agents to take work off your plate.

Create agents for customer support, sales, marketing, engineering, ops, and more. Give each one a role, connect the tools they need, and hand off the work.

Your agents can work on their own or together, sharing context and handing off tasks to get bigger jobs done.

Start with one agent. Build a whole team around you.

Make your first agent

Evolving AI: Amodei's plan to slow AI reads in Beijing as a plan to keep China behind.

China's pushback answers Anthropic CEO Dario Amodei's essay, We Must Pace the Frontier, which warned that a Chinese lead in AI would endanger the United States and the world. China's state press read his asks as containment, pointing at the ban on advanced chip sales and the call for tighter security inside American labs. Its editors argued Washington wants the gap locked in first, with talks on a shared slowdown coming only after that.

Washington and Moscow signed their nuclear limits only after each side held a lead it could live with, and Beijing sees that same order running underneath the pacing plan. American agencies and AI labs have set out proof that Chinese models distilled capability from Claude and GPT, so a serious charge now sits on the table from the other direction too. What comes next rests on whether two rivals this deep into a race can trust each other at all, and on how fast the AI risk itself gets real attention. Neither one survives without the other, since a deal nobody trusts goes unsigned and the safety work behind it stays on paper.

Evolving AI: SpaceXAI is going to build a company from scratch using Grok Bot without a plan.

Grok Bot works as a set of agents on cloud machines with a browser and a terminal, so it can write code and carry multi-step tasks while the three of them approve whatever gets built. One idea has to become a business plan and then a product people can use before Thursday evening. The bot is still in beta and comes bundled with a SuperGrok subscription, so the same setup is available to buy.

Register for the GrokBot Session

SpaceXAI is betting its own name on Grok Bot running every function of a company end to end, live, with nothing edited out. The market has watched plenty of agent launches land as a short demo clip filmed under friendly conditions. A three-day build that has to survive real deadlines says something firmer about where this kind of work has got to. Anyone running a business can watch the handoffs and work out how much of their own week software like this could take over.

QUICK HITS

⛓️ A Senate bill would give AI developers 20 years in prison for building superintelligence, matching nuclear weapons penalties.

🔁 A researcher released the Recurrent Looped Transformer, claiming transformers with infinite reasoning depth.

📈 Anthropic told investors it will be profitable for a second straight quarter.

🏛️ Obama told Democrats AI should be a central agenda item, saying it's what he'd run on in 2028 and that he's neither an accelerationist nor a doomer.

 *partner link