In partnership with
Hey {{first_name|Investor}} -
On Saturday, September 12, Dario Amodei, the CEO of Anthropic (the company behind the AI assistant Claude), published an essay called "We Must Pace the Frontier." His argument: the industry should slow down how fast AI models get more capable, to give safety work time to catch up.
Then something unusual happened. His competitors agreed. Sam Altman of OpenAI wrote, "I agree with Dario that we need to pace the frontier," and committed OpenAI to the part of the plan that puts outside safety evaluators inside the labs. Elon Musk wrote, "Dario is right."
On Monday, September 14, chip stocks fell. Nvidia dropped 3% and Intel 6% (Los Angeles Times).
So the market read it as a slowdown. I don't think the building slows, and sincerity has nothing to do with it. It's a problem that shows up any time rivals promise each other restraint.
The two suspects
Picture two people arrested for the same crime and questioned in separate rooms. Neither knows what the other is saying.
If both stay quiet, the police have a weak case and both get a light sentence. That's the best outcome for the pair.
But look at it from inside one room. If the other person stays quiet and you talk, you walk. If the other person talks and you stayed quiet, you take the whole sentence. Either way, talking looks better for you.
So both talk. Both end up worse off than if they'd both kept quiet.
That's the prisoner's dilemma. Everyone would win by cooperating, and each individual is better off defecting, so cooperation falls apart unless something outside the room holds it together.
The same shape, with AI labs
Swap the suspects for AI companies, or for countries.
Everyone is better off if all of them slow down: more time for safety, less chance of an accident that lands on the whole industry. But each one is better off racing if it can't be sure the others really stopped. The one that keeps pushing gets the better model, the customers and the lead.
Amodei's essay says this out loud. He writes that pacing "will be limited by the lead that US companies have over authoritarian regimes, chiefly the Chinese Communist Party," and that slowing by more than that lead lets those projects pull ahead. In plain terms: we can only slow down as much as our head start allows.
He also names the hard part. Any international agreement, he writes, "must either have ironclad verifiability, or must be limited enough that defection would not be militarily existential." Defection is the game-theory word for cheating. And the difficulty, he adds, is confirming that neither side is holding secret models it doesn't test in public.
That's the dilemma in the author's own words, in the essay arguing for restraint.
What breaks a dilemma
Game theory says cooperation holds when three things are true, and they're worth memorizing because they apply far beyond AI.
Someone can verify. If I can check whether you kept your word, your promise means something.
Defecting costs you. A penalty, a lost contract, a broken relationship worth more than the one-time gain.
The game repeats. People who expect to deal with each other again behave better than people meeting once.
Amodei's first step is aimed straight at the first condition: outside evaluators with employee-level access inside the labs, so an outsider can see what's actually being trained. Altman said OpenAI would do the same. That's real, and it's the part of the plan that could work, because it's the part that creates verification.
The second and third conditions are missing between countries. There's no penalty for a government that quietly keeps building, and no referee.
Washington answered the next day
On Sunday, September 13, President Trump told reporters, "We're leading China in AI... whoever wins AI wins." He described the safety warnings as overdone.
This newsletter doesn't take sides on politics, and the structural point stands on its own. The pledges are voluntary. Washington's stated priority is staying ahead. And no one has proposed a mechanism that would make slowing down cheaper than racing.
What this means for your money
Separate what people say from what gets built. That's the whole issue in one line.
Pledges are cheap and fast. Buildings are slow and expensive. A data center under construction today was financed years ago, and it runs whatever models exist when it opens. Contracts for its electricity run 10 or 20 years.
So when you see a headline about an industry agreeing to restrain itself, the useful question is simple: does anything in the agreement change what gets spent? In this case, the concrete commitment is letting outside evaluators watch the training. That changes oversight, and it doesn't change a single construction schedule.
Which is why I read Monday's drop as sentiment. Demand didn't change that day, and sentiment moves like it are invisible a year later. If a slowdown ever shows up in orders and capital spending, that's the thing to act on, and you'd see it in the quarterly numbers before you saw it in an essay.
Three questions for the next "industry agrees to" headline
Who verifies it? If the answer is "they said so," it's a statement of intent.
What does cheating cost? If nothing, expect cheating.
Does it change spending? Follow the capital budget. Press releases are free.
Two suspects in separate rooms both talk, every time, unless someone can see into both rooms.
Make Your Own Alpha is live. Today's issue handed you one habit: separate what companies say from what they spend. The book teaches the full set of frameworks I use to decide what earns a place in my portfolio and what has to prove itself first. Get your copy here.
Stay discipined, Koh
P.S. A word from this post’s sponsor.
These 7 Stocks Are Built to Outlast the Market
Some stocks are built for a quarter… others for a lifetime.
Our 7 Stocks to Buy and Hold Forever report reveals companies with the strength to deliver year after year - through recessions, rate hikes, and even the next crash.
One is a tech leader with a 15% payout ratio - leaving decades of room for dividend growth.
Another is a utility that’s paid every quarter for 96 years straight.
And that’s not all - we’ve included 5 more companies that treat payouts as high priority.
These are the stocks that anchor portfolios and keep paying.
You can download this report for free as of today, but it won’t be free forever.
This is your chance to see all 7 names and tickers - from a consumer staples powerhouse with 20 years of outperformance to a healthcare leader with 61 years of payout hikes.
Disclaimer: Nothing in this email, the Reset and Invest newsletter, the Make Your Own Alpha book, or any course or digital product from Starshine Media LLC constitutes investment advice or a recommendation to buy or sell any security. Numbers and observations are as of publication. I may hold positions in companies discussed. Always do your own research and consult a licensed financial advisor before making investment decisions.