2025-07-18
minor edit 2026-02-13
How to enforce an AI ban
Disclaimer
Why this post
- I was recently asked by someone what enforcing a ban on ASI actually looks like. I am not confident in many of the details below but here's a rough sketch. I am most sympathetic to the recommendations from the last few paragraphs of Yudkowsky's TIME article. Since most of these plans make sense to execute only once a significant number of people are convinced on x-risk, I would prefer putting effort into persuading people of the risk, than in fleshing out all the details of these plans. I do think someone should flesh out the details.
Main
- Assume govts of US, UK, China and Israel agreed that preventing human extinction from ASI is a priority equal if not greater than preventing human extinction from nuclear war. Assume that significant fraction of population of each country was also convinced about the same.
Then govts can do the following:
- Shut down all semicondutor fabs for GPUs (not CPUs)
- Confiscate all GPUs from existing large labs.
- Use surveillance data they already have, to track down hundreds of smaller buyers and confiscate those too.
- Pass a law making GPU ownership and sale a punishable offence, with higher penalty for higher volume (similar to Schedule A drugs)
- Ideally melt all the GPUs. Second best option is to rent them out for non-AI use cases in a transparent manner.
- Passing a law banning discussion of AI research ideas. Imprison people who are found to openly violate these laws. (Just as say, people discussing industrial production of Schedule A drugs also end up in prison.)
Since everyone is unlikely to voluntarily agree to ASI x-risk being number one priority, pulling this off in the real world will be difficult. There are multiple political moves that may have to be undertaken by various actors.
- Persuasion targeted at various actors on why ASI x-risk is the number one priority.
- Elites within govts may have to threaten war on elites within other govts that don't support a ban on ASI, including possibly escalating to threats of nuclear war.
- Elites with govts may have to livestream as many of their actions as possible, to gain trust and prove they're not building a secret rogue datacentre for themselves. They will have to prove this to elites in foreign govts and to their own citizens.
- People may need to demand accountability from their govt to ensure this happens, be it via electing new leaders, mass protest, supporting whistleblowers and cyberhackers to force transparency, and so on.
- Some people will need to take on these special roles for political action, like actually running for election with AI x-risk as an agenda, or actually becoming a whistleblower.
- Break down of law and order is a live possibility in many countries, especially countries that are deemed irrelevant in the AI race. AI will lead to drastic change in economic circumstances, which in turn can lead to political circumstances such as regime change. I do not have an easy solution to this.
- There are some (hopefully unlikely) worst case worlds where break down of law and order occurs and yet all the GPUs have not been confiscated. Civil war may occur and citizens may have to volunteer to join either side's militia to ensure all GPUs are confiscated and melted.
Subscribe
Enter "Subscribe: " followed by Signal username, email address or whatsapp phone number. You will receive atmost one update per month. View previous updates here.
Alternatively, you can use RSS, or a third-party service like blogtrottr.
Comment
Enter "Comment: " followed by comment. Alternatively, contact me