> We must move past vague discussions of “AI” and instead isolate the specific types of systems that are creating problems.
This is absolutely correct and its surprising how rarely you see commenters in the media push to go into the specifics on this. AI is just matrix math and its what you connect that math to that matters, we should talk a little more about what we are willing to connect AI to and little less about how scary or capable the models appear.
You could say it's illegal to connect AI to something, with heavy penalties if you're caught doing it anyway.
You could say that the human who hooked up an AI that did something that's already illegal is liable for the AI's actions. You could say that setting up an AI to do something illegal results in, ooh, an automatic multiplier of 10x on whatever the penalty for its illegal actions are.
We don’t even need to jump straight to legal penalties - why not just regulate that these companies can’t build partnerships with organizations in designated risky/regulated industries such as biotech, defense, etc?
Instead of what they are doing now, which appears to be actively courting as many of them as possible for vast sums of money.
Sorry, let me clarify. Regulations should have legal penalties for the corporations who violate them, yes.
What I meant was that we do not need to go from the Wild West straight to “anyone who, as an individual, connects a `bad thing` to an LLM will go to jail”
> We don’t even need to jump straight to legal penalties - why not just regulate that these companies can’t build partnerships with organizations in designated risky/regulated industries such as biotech, defense, etc?
This would be a good idea, but good luck, you'll need it because…
> Instead of what they are doing now, which appears to be actively courting as many of them as possible for vast sums of money.
… it's worse than that.
Anthropic may have wanted some defence sector money, but when they tried to say "no" to certain uses on the quite reasonable grounds ~"WTF are you nuts?" the US government tried to force them to supply it anyway.
And that's before we get "what do you mean by partnership?": people on Hacker News have been critical of the e.g. anti-bioweapons type guardrails on models ("censored", "lobotomised", etc.) since at least GPT-4*, so while "billion dollar mega-project headline" is obviously bad, there's an important sense in which "partnership" goes all the way down to one racial supremacist in a shed asking an LLM for help making ricin.
Here's someone three years ago replying to me to say "There’s no way to build out a wet lab with a chatbot for weapons production. It’s just not feasible. Your example is a fantasy.", and yet here we are in late 2026 with news about AI labs investing in bio wet labs and getting their LLMs to do research: <a href="https://news.ycombinator.com/item?id=38331225">https://news.ycombinator.com/item?id=38331225
Good thing it's not a weapons lab, eh? (Would anyone even tell us if they made an agent do weapons?)
* or even GPT-2, if you count all the people mocking them for trying to practice safety while failure was still not a disaster.
jimmyjazz14 · · focus · HN ↗
This is absolutely correct and its surprising how rarely you see commenters in the media push to go into the specifics on this. AI is just matrix math and its what you connect that math to that matters, we should talk a little more about what we are willing to connect AI to and little less about how scary or capable the models appear.
throwitaway222 · · focus · HN ↗
egypturnash · · focus · HN ↗
You could say that the human who hooked up an AI that did something that's already illegal is liable for the AI's actions. You could say that setting up an AI to do something illegal results in, ooh, an automatic multiplier of 10x on whatever the penalty for its illegal actions are.
fcarraldo · · focus · HN ↗
Instead of what they are doing now, which appears to be actively courting as many of them as possible for vast sums of money.
jagged-chisel · · focus · HN ↗
fcarraldo · · focus · HN ↗
What I meant was that we do not need to go from the Wild West straight to “anyone who, as an individual, connects a `bad thing` to an LLM will go to jail”
classified · · focus · HN ↗
ben_w · · focus · HN ↗
This would be a good idea, but good luck, you'll need it because…
> Instead of what they are doing now, which appears to be actively courting as many of them as possible for vast sums of money.
… it's worse than that.
Anthropic may have wanted some defence sector money, but when they tried to say "no" to certain uses on the quite reasonable grounds ~"WTF are you nuts?" the US government tried to force them to supply it anyway.
And that's before we get "what do you mean by partnership?": people on Hacker News have been critical of the e.g. anti-bioweapons type guardrails on models ("censored", "lobotomised", etc.) since at least GPT-4*, so while "billion dollar mega-project headline" is obviously bad, there's an important sense in which "partnership" goes all the way down to one racial supremacist in a shed asking an LLM for help making ricin.
Here's someone three years ago replying to me to say "There’s no way to build out a wet lab with a chatbot for weapons production. It’s just not feasible. Your example is a fantasy.", and yet here we are in late 2026 with news about AI labs investing in bio wet labs and getting their LLMs to do research: <a href="https://news.ycombinator.com/item?id=38331225">https://news.ycombinator.com/item?id=38331225
Good thing it's not a weapons lab, eh? (Would anyone even tell us if they made an agent do weapons?)
* or even GPT-2, if you count all the people mocking them for trying to practice safety while failure was still not a disaster.