WR 53 (part 2): US Open-Source AI Governance
Part 2 of the Weekly Reckoning for the week of 10/2/25
Welcome back to the Ethical Reckoner. As promised, here’s Part 2 of this week’s newsletter: the Extra Reckoning on my new report published with the Center for AI Policy, “US Open-Source AI Governance Balancing Ideological and Geopolitical Considerations with China Competition.”
This edition of the WR is brought to you by… Scam, Inc.
Looking for the Reckonnaisance? Find part 1 here:
Extra Reckoning
Last month, a Chinese company called DeepSeek released an artificial intelligence model that matched—and in some ways beat—the best American AI systems from companies like OpenAI and Google. What made this particularly interesting? DeepSeek’s code is open-source, meaning anyone can download and use it, and it trained with far fewer resources than those closed models (although it benefitted from their advancements).
This has sparked heated debate in Washington. Some lawmakers are calling for strict export controls on open-source AI to prevent China from using American technology. Others argue that such restrictions would backfire, hurting American innovation while doing little to slow China's progress.
Our new report examines this debate and finds that broad restrictions would likely do more harm than good. Here's why:
Unlike physical technology like computer chips (which have been subject to extensive export controls), AI code is just information, so once it’s out there, it’s basically impossible to contain.
Open-source collaboration has been crucial for developing beneficial AI applications in healthcare, science, and education, so preventing its use would be really bad in general.
Restricting American open-source AI could actually accelerate China's push for technological independence and increase its global influence.
We also analyze the performance of models in both Chinese and English. TL;DR: US closed models are still the best, but Chinese closed models are catching up. The EU is languishing in the middle.
But Chinese model advances are likely to slow as chip sanctions bike, and proprietary models will probably leap forward when they adopt the innovations from DeepSeek and other open models.
Instead of broad restrictions, we recommends more targeted approaches:
Requiring security assessments before releasing powerful models
Creating public databases of security audits
Investing in technical safety measures
The bottom line: While China's AI advancement deserves serious attention, destroying the open-source ecosystem that has helped make America a leader in AI would be counterproductive. We need smart, targeted policies that protect security while preserving innovation.
Want to learn more? Read the executive summary here. Want to read even more? Read the full report here.
I Reckon…
that if you like not being scammed, you should be worried about the siege of the Consumer Financial Protection Bureau.
Thumbnail generated by ChatGPT with the prompt “Create a brushy abstract Impressionist painting representing the concept of technological US-China competition. The background should look like a circuitboard”.