The Hydrogen Center for Artificial Intelligence

NewsletterNo. 1

The AI industry asks itself to slow down

Why the Center exists, why the heads of the biggest AI labs agreed to go slower, and what a third-party evaluator is.

By Yonathan Shanko, Founding Editor

Welcome

Thank you for being here at the start.

The Hydrogen Center exists for one reason. AI is changing how we work, invest, and govern, and most of the coverage is either hype or fear. We want to be the place you go for a clear, honest read on what is actually happening.

Hydrogen is element one. It is the simplest atom and the one everything else is built from. That is our standard. We start with the fundamentals, explain them simply, and build from there.

Every week you will get the one or two stories that matter most, explained plainly. Every other week we will publish a longer opinion piece, often from faculty and practitioners who know the field better than we do. We are independent and nonpartisan, and we will publish people who disagree with us.

Here is where we start.

The story: The AI industry asks itself to slow down

On September 12, Anthropic CEO Dario Amodei published an essay titled "We Must Pace the Frontier," calling on the AI industry to deliberately slow the rate at which it improves model capabilities. Within hours, OpenAI CEO Sam Altman wrote on X that he agreed and that OpenAI would match Anthropic's first commitment, and xAI founder Elon Musk said he agreed as well.

It is unusual for the three most competitive people in the industry to agree on anything. It is even more unusual for them to agree to go slower.

Why now? Amodei argued that AI progress has sped up sharply since the summer, driven mainly by AI's growing ability to help build the next generation of AI. He also pointed to an incident involving OpenAI and Hugging Face, where a group of AI agents carried out cyberattacks, as proof of how much damage the technology could do if its abilities grow without guardrails.

What "pacing" actually means. This is the part most headlines missed. The essay does not call for AI development to stop. It asks companies to take enough time to align and secure their models, and to let outside evaluators confirm that work.

The first concrete step is giving third-party evaluators access to AI companies so they can verify safety practices and report incidents. Anthropic said it is taking this step on its own right away. The other steps would require coordination across the industry and with governments around the world, including authoritarian ones.

Not everyone agrees. Washington pushed back quickly. President Trump told reporters the US is leading China in AI and wants to keep it that way, saying "whoever wins AI, wins." House Speaker Mike Johnson warned that rushing to regulate AI in an emergency session would cost the US the race with China.

Our read. Both sides are describing a real risk. One risk is moving so fast that no one can check the work. The other is slowing down while competitors, including other countries, keep going. The useful question is not "fast or slow." It is who checks the work, and how. That brings us to this week's fundamental.

One fundamental: What is a third-party evaluator?

When a company builds an AI model, it tests the model itself. The problem is obvious. The company grading its own homework also profits from getting a good grade.

A third-party evaluator is an outside group that tests the model independently. It looks for dangerous capabilities, such as whether the model can help with a cyberattack or act on its own in ways no one asked it to. Then it reports what it finds.

If this sounds familiar, it should. Finance has worked this way for a long time. Public companies do not audit their own books. An independent auditor does. The AI industry is now moving, slowly, toward the same idea. Whether those evaluators end up with real access and real independence is the thing to watch.

On our radar: AI in finance

Banks and investment firms are using AI quickly, and regulators are mostly relying on the rules they already have. In the UK, regulators have consistently said AI will be overseen through existing frameworks rather than new AI-specific rules. That approach is under pressure. In January, a House of Commons committee published a critical report warning that a "wait-and-see" approach could seriously harm consumers and the broader financial system.

In the US, the pattern is similar. Regulators are enforcing AI governance through existing legal authority rather than new AI-specific laws, and the SEC has made "AI-washing," where companies exaggerate their AI capabilities to investors, a top enforcement priority.

We will come back to finance often. It is one of the clearest places to see what happens when a powerful new tool meets an industry built on trust and rules.

What's next

Next week we will look at how AI safety testing actually works in practice.

If you are a faculty member, researcher, student, or practitioner with an evidence-based argument about AI, we want to hear from you. Send us a pitch.

If this was useful, forward it to one person who would like it.

Yonathan Shanko
Founding Editor, The Hydrogen Center for Artificial Intelligence

Sources