Should the US govt’t have fewer AI rules to promote innovation, even if risks increase? Or should it have stricter AI safety rules, even if it slows innovation?

Stricter rules (71%) – "We should establish stricter guidelines on AI to enforce security and measured implementation but also work with global partners instead of through competition. The US is on an island right now and that is not realistic in this global environment. We need to shift this isolationist approach."

• "Just like anything else, AI needs to be applied in the correct way. The challenge would be finding regulations that seriously reduce risks without slowing beneficial innovation. If that does not happen we run the risks of unprotected privacy & personal data, discrimination & bias can increase compromising important decisions like hiring or lending & there could be an increase of risks of capable AI, such as misuse for cyberattacks or the creation of convincing misinformation. It’s scary when AI is being abused to create images that appear real but in fact are not."

"Our businesses, critical infrastructure, utilities and banking are are all interconnected - with very few safeguards or redundancies built in. The risks of unintended consequences are far higher than the (so far) limited rewards. There are simply too many "unknown unknowns" at this point. If the unofficial motto were not "Move fast and break things," I would feel much better about it."

• "I think we've seen the dismal results of the lack of regulation by the federal government on previous systems.... especially social media. Allowing that has been nigh on criminal. To take the same laissez fair attitude towards AI could be catastrophic."

"The US should have stricter rules on AI use to insure public safety; even at the cost of slower innovation and competitiveness. Freely releasing newer AI models to the public has a major issue: increased risk of jailbreaking and accelerated cyberattacks; used for bypassing rules and leaking data. Although Zhipu’s open- weight model is a cheaper alternative for storing data-without the safeguards-it increases the risk of hackers swiping private informations. This will quickly become a wide ranging problem; making companies unreliable and unsafe-creating ethical and societal problems. Consequently, in order to gain people’s trust and protect public safety the government should enforce security laws for company use of AI."

• "AI has already responded to some individuals with harmful advice. That needs to be prevented and the only likely method currently available is via legislation. The AI companies do not appear terribly interested in reining in their creations."

"Let me be straight with you. AI is more regulated and adopted in China than it is in the US. Few US companies are doing open weights. Prices are going up, and that includes the price per token (looking at you fable 5). The reason US companies aren't as far ahead anymore isn't because of regulation, it's because of strategy. I do think AI needs good regulation. I don't know about the mythos ban. I think Anthropic got it's deserts with its "AI is dangerous and will replace you; isn't that really exciting?" Strategy with that one. We've very likely got enough data centers to continue innovation if US companies were willing to do something like license their models with a subscription to run locally. That's not the American way of doing business. All that said GLM 5.2 is mostly hype. Benchmarks are hype. It's only about as good as opus chat GPTs 4o research on a very good day when it comes to code. It's nowhere near Fable 5, much less Mythos. Though the gap is smaller than it used to be."

Fewer rules (18%) – "While inhibiting access to new models doesn't stop innovation, it does put us at risk of attack and grants a select group of companies an unfair advantage. Companies with access can find and fix their vulnerabilities but they can also use it to create new capabilities faster and better, at lower cost, than those without access. I applaude Microsoft for providing access to open weight models. That helps us all have a level playing field. At least until China begins to suppress the release of those models. Then we know the trouble is about to really begin."

• "Still should be rules and some caution but certainly "fewer" rules than stopping the release of "models that are good at cybersecurity"."

Unsure/other (11%) – "Stricter vs fewer rules is the wrong question, because rules aren't categorized just on a binary strict or not strict rubric. I do think AI needs some rules and structures around it, but we need to have a clear goal in mind so we can also rate federal laws on effectiveness to ensure we're not creating rules just for the sake of having rules. I don't want the government meddling just to meddle because I do belive in the power of a free market, but we also need some oversight on this immensely powerful and potentially dangerous creation. TLDR: We need both our rules to be both few AND strict when it comes to AI."