How do we turn AI into a source of human flourishing when two rival great powers are developing it at the same time?
AI is helping people cause harm and helping other people prevent it. Meanwhile, the United States and China are competing over the infrastructure, companies, and capabilities that will shape its development.
In my YouTube conversation with Alvin Graylin, we explored China, education, and the possibility of abundance. Here I want to work out what principles could make that abundance real in family, community, and business life, while reducing the dangers of a technology that makes bad actors more capable too.
I want to connect the questions facing American and Chinese negotiators with the choices we make in companies, schools, and homes. The question throughout is whether our response makes people safer and gives them more opportunity to improve their lives.
On September 12, Anthropic CEO Dario Amodei called for slowing advances in the most capable AI systems so that safety work can keep up. His essay, “We Must Pace the Frontier,” proposes outside evaluators, coordination among democratic countries, and international cooperation.
Sam Altman committed OpenAI to the outside-evaluator measure too, with details to follow. That leaves open how the reviews will operate and whether companies will agree to slow development.
On September 10, Anthropic reported that a Yemen-based engineering cell had used Claude to help develop guided weapons software. Its account didn’t establish that a working weapon had been fielded. It also described AI-assisted surveillance targeting dissidents. Those are serious uses of a technology that makes people more capable, including people whose purposes are destructive.
Also this week, American security researchers at Calif disclosed a vulnerability they had found in WeChat with AI assistance, demonstrated an attack that could spread between accounts, and worked with Tencent to get it fixed. This was a controlled demonstration, not evidence of a global outbreak. The researchers reported that Tencent had mitigated the issue before publication.
Reuters reported preparations for AI safety talks ahead of an expected September 24 Trump–Xi summit, while a White House official disputed that an AI meeting was planned for mid-September.
The question I want to bring to those talks is this: how will the American and Chinese sides understand their respective needs for safety after these incidents, while each remains worried about the other’s power? And can they turn that concern into arrangements that help people in both countries flourish?
I agree with Alvin on a fundamental point: great-power competition increases the net risk. Governments that believe they must win at almost any cost have reasons to move faster, share less, and interpret another country’s progress as a threat. That changes the conditions under which everyone develops and deploys these systems.
If we believe AI can produce abundance, we need to take seriously what it could do for daily family and community life. Over the last year, I have seen AI become a massive accessibility aid in my family. I talk a lot about business. I care deeply about helping people build companies and do better work. That is an enormous part of the opportunity. It is also only part of it.
We have a chance over the next few years to influence how people use AI and who benefits.
This briefing covers:
Why there is no finish line. Nobody won the electricity race, and the research on general-purpose technologies says why that analogy holds.
Innovation and misconduct, separated. How to take Anthropic’s allegations against Chinese labs seriously without concluding those models are Claude underneath.
What abundance looks like at home. The accessibility case, and why improvements at work are too small a measure of it.
Making everyday AI affordable and usable. What a 280-fold price collapse does and doesn’t tell us about access.
Room for the people closest to the problem. Where entrepreneurship fits, and what the education research says about AI that hands over answers.
Goals, limits, and what the summit could verify. The agent incidents worth taking seriously, and how I’d judge a US–China safety agreement.
