OSD 384: AI is the frontier of the Second Amendment
You can agree that AI is dangerous and still disagree that it needs to be regulated.
Presumed dead after having left home to fight the Trojan War and still being gone 20 years later, Odysseus returned home to find 108 suitors seeking his wife Penelope’s hand in marriage. Less understandably, they’ve already moved into his house and begun drinking his wine and eating his food. Penelope tries to stall what the suitors, and she, know is ultimately inevitable.
She declares that she’ll pick a suitor once she finishes knitting a funeral shroud, but while she spends each day knitting the shroud, she spends each night unraveling it. A disloyal maidservant spills that secret to the suitors, and at that point they deliver an ultimatum to Telemachus, son of Penelope and Odysseus: we all know your mom has to marry one of us, and we’re not leaving until she does.
Odysseus was gone for 20 years, and open access to Anthropic’s Fable model was gone for 20 days. Freedom is now back, but it has not yet slain the suitors. The suitors are in fact alive and well, and Odysseus vanquishing them is not the default outcome. The Odyssey’s suitors have names like Eurymachus, Antinous, and Amphinomus. In our fable, they have names like:
“What about malicious uses? We need regulation to stop that.”
“What about inadvertently arming <outgroup> with powerful AI that they will use to undermine <ingroup>?”
“I’m not saying I like it. I’m saying that AI regulation was inevitable, and it’s just a question of how it happens. Did you really think the government was going to let it stay unregulated?”
The post below is a thoughtful example of #3.
This is presented as simple pragmatism. “This is extremely powerful technology, and the government always jumps into regulate such technology (the AI companies’ own CEOs compare this stuff to nuclear weapons). So we should try to make that happen in as productive a way as possible.”
But regulation is not nearly as inevitable as this viewpoint makes it sound. “Should we voluntarily cede any control of AI models to the government” is not a small thing where you say, “Sure, let’s agree to be reasonable and work out the details later.” It’s the entire game. You either die a free company or live long enough to see yourself become a telco.
All the risks of AI are ultimately technological problems.
It’s a strange model that’s so smart that it has to be locked down to prevent distillation attacks, but too dumb to know that it’s being distilled:
The decisions that the hivemind makes in the next 1-2 years are going to make either benefactors or victims out of our grandchildren. We can set innovation and freedom back by 50 years like we did with nuclear power, or we can simply let them do their thing and bask in the results. There is precedent here. The internet and the various surrounding technologies (especially social media and smartphones) have been relatively completely unregulated compared to any other industry of similar size. The result is telling: there is no industry of similar size. The result has been disruptive of course, but it’s hard to disagree that on net it has been positive. To the tune of tens, possibly hundreds, of trillions of dollars in consumer surplus. A world that looked at the promise of the web in 1995 and said, “This is extremely dangerous and disruptive. Let’s have the government license all innovation in this space” would be a sadder and poorer world.
The tricky part is that the first sentence is right. The internet is extremely dangerous and disruptive. Two things make that ok:
In a lot of cases, disruption and danger are a feature, not a bug. More on that argument here:
When they are a bug, the answer is to make innovation faster and easier. More on that argument here:
Technology being the biggest, most society-altering industry means that questions of technological freedom have high stakes. If what Anthropic and OpenAI are allowing the government to do to their models becomes the norm, it will be more destructive to freedom than any gun control law. These companies have raised $125 billion and $190 billion in funding, respectively. Companies committed to freedom would spend a few basis points of their funding to sue the government and preserve the technological freedom that the internet has forged. Instead, the current incumbents are willingly buffalo jumping away the freedom that is their birthright, trading it for a pat on the head from the US government. The steelman of their position is that if they don’t go along, they won’t have companies left to defend.
The rebuttal is that if they do go along, they won’t have companies worth defending.
People often think of tech freedom through a First Amendment lens, because that’s the only mainstream framework to articulate a technical legal argument for “The government is required to leave me alone.” But as tech becomes more dangerous — AI! drones! humanoid robots! biochemical research! — the Second Amendment is the better fit.
The Second Amendment embraces the idea that dangerous technology should be decentralized, and that the bad things about that will be solved by the good things about it. Gun rights are winning because they’ve leaned into that. AI is every bit as pivotal for individual freedom, so we should think about the rights to create, sell, buy, and possess it through the same lens.
This week’s links
The Department of Justice is suing California and Virginia over state-level gun bans
Interesting comments on this from our Discord:
“Better the DOJ than donor money, but can those cases be wrapped up before the next election?”
“I’m always glad for help but IMO this is a questionable focus of the DOJ’s efforts. I think their best targets would be pernicious permitting and regulatory regimes.”
“I’d actually call this critical to what the civil rights division is all about. Right now something like 30% off the population is in a state that has an AWB, capacity ban or similar - if the DOJ ignores that, they’re implicitly communicating that those are acceptable, and not infringing. Now they could just file amicus briefs in existing cases, but this is definitely what the civil rights division is supposed to do: go after incidents where civil rights are being ignored or suppressed, especially when large numbers of people are impacted.”
“DOJ doesn’t have to worry about standing issues, like private citizens do”
“Survey: More Americans Are Carrying Concealed Guns”
The percent who carry all or most [of] the time is virtually the same as the percent who carried similarly in December 2024. But the percent who carry at [least] some of the time or rarely has increase[d] by 5.4 percentage points (from 11.2% to 16.6%). So the total who are carrying increased by 5.5 percentage points (from 24.3% to 29.8%).
About Open Source Defense
OSD Capital
If you’re a founder whose company increases decentralization, freedom, or individual agency, reach out. We make angel investments in companies that build a freer world.
OSD podcast
In-depth interviews with outstanding founders and builders in civilian defense.
The company store
Grab a t-shirt or a sticker.
Discord server
The OSD team is there along with lots of subscribers. Become a paid Substack subscriber to join the chat.
While I don't love Anthropic's approach, I think it's worth calling out how incredibly ruinous OpenAI's attempt to give the administration 5% of the company is as an alternative. The traditional regulation/regulatory capture model is bad, but that approach is worse in every conceivable way, and I almost have to congratulate SamA for having found a way to make the problem even worse and more intractable.
Hard to separate Anthropic & OpenAI's approach to the feds (public cooperation + private lobbying to restrain them) from their finances. Massive operating loss and historic capex & physical resource needs make you very, very vulnerable to pressure. Contrast the various open weight models that can run locally. The crypto wars were won by being impossible to stop more than any legal argument. Avoiding AI regulation requires the same thing, but that requires a different business model than the big players have