I don't think "bans don't work" is the correct take-away from the USA's "war on drugs".

Maybe it would be fair to say "you can't just ban something." For example, tobacco-sales are restricted to minors in the US. And smoking among minors did drop sharply (until vapes came along). There were several controls used: outlawing sales, raising overall prices (so that teens generally couldn't afford to smoke), education campaigns, and making smoking more difficult in general.

In contrast, the War on Drugs mostly involved incarcerating vulnerable drug-users. Some similarities to the current war on immigrants.

I think a tobacco-style attempt to ban AI would look like: - A large electricity surcharge to inference-providers, who would pass the cost to customers, making AI far more expensive. Maybe it would be worthwhile to drop $10M/year on code-assistance or slide-deck clip-art, but not for an individual to ask how to restore their Instagram account. - An education program telling people that they sound cringe when they write in AI voice, and which contexts it's appropriate/inappropriate to do that. (Don't AI your wedding vows.) - Certain types of harm would be the responsibility of the provider. Maybe if the AI medical assistant fills out the form incorrectly then the provider is liable for the malpractice suit. Or if it tells a teen to kill themselves the provider is liable for wrongful-death. These would likely result in the provider trying to cut off access to any of the groups where they couldn't shift liability.