Sep 25, 2026 · 1h 4m
Listen from 17:40
Listen at 17:40
States test tougher rules for AI and data centers
How To Actually Regulate AI — ft. Alex Bores
The episode examines whether governments can contain catastrophic AI risks and make data centers bear the environmental and infrastructure costs they create.
- 1Independent audits only work when companies cannot select or control the auditors overseeing their systems.
- 2AI liability remains unsettled, especially when harm reflects developer misalignment rather than deliberate misuse by users.
- 3Data centers are turning abstract AI policy into local fights over energy, water, pollution, taxes, and economic inequality.
Don't miss
Alex Bores explains why data centers have become the first place voters directly encounter AI policy, through local permitting fights over energy, water, pollution, and costs.
The brief
Alex Bores joins Ed Elson and Scott Galloway to argue that AI policy must preserve its potential benefits while confronting catastrophic risks, weak accountability, and the political costs of rapid deployment.
Bores points to independent audits, deployment monitoring, and outside evaluator access as starting points, while warning that company-selected auditors would turn oversight into self-regulation.
The liability debate turns on a crucial distinction: users may misuse AI, but developers can also create systems whose misalignment causes harm without clear legal consequences.
Data centers bring AI policy into everyday politics through local permits, exposing disputes over energy, water, pollution, inequality, and whether operators should pay their full infrastructure costs.
New York, Illinois, and California offer competing state-level experiments, suggesting meaningful intervention remains possible even as AI development accelerates.
What was said on this episode
27 statements · 13 positive · 8 negative · 2 mixed · 4 neutral
Manhattan will be viewed as having experienced a golden age in 20 years.
“I think in 20 years, we're going to look back on Manhattan and think that was a golden age.”
Listen at 2:53
Bores passed 35 bills, approximately five concerning AI and technology.
“I've passed 35 bills in my four years. I think maybe five of them had to do with AI and technology.”
Listen at 8:44
U.S. states will become highly active on AI regulation next year.
“the states are going to be really active next year”
Listen at 14:21
Governments could require monitoring of every AI deployment.
“You could require monitoring of every AI deployment.”
Listen at 15:10
AI developers should mandatorily report safety incidents.
“there should be mandatory reporting of safety incidents”
Listen at 15:47
AI governance decisions should not be left to five Silicon Valley billionaires.
“We can't leave those decisions up to five Silicon Valley billionaires.”
Listen at 19:43
Existing laws are insufficient to address the Hugging Face attack.
“I don't think we have all the laws on the books”
Listen at 24:55
A preliminary ruling assigned responsibility for an AI agent’s action to its user.
“The early preliminary ruling was that it's the user of the software.”
Listen at 25:44
Users should bear responsibility for misuse, while developers bear responsibility for misalignment.
“you would have misuse, the user being responsible for, and misalignment, the developer being responsible for.”
Listen at 26:43
Self-replicating AI agents could become impossible to shut down.
“if these agents were to make copies of themselves elsewhere, such that no one could shut it down”
Listen at 28:21
State attorneys general have incentives to hold AI actors accountable.
“I think there is some incentive to hold people accountable here.”
Listen at 30:32
Slowing U.S. AI progress would substantially slow China’s AI progress.
“if we slow down in the US, that will also slow down China, not 100%, but a substantial amount.”
Listen at 33:28
AI systems currently lead 26% of OpenAI projects.
“Right now, they lead 26% of the projects within OpenAI.”
Listen at 36:14
At full AI-led project development, internal processes will become harder to understand.
“when that gets to 100%, it just becomes harder to know what's happening on the inside until it's too late.”
Listen at 36:18
Recursive self-improvement should be prohibited without internal understanding.
“you should never do recursive self-improvement if you're not going to know what's on the inside”
Listen at 36:27
Clear, unbiased AI standards can improve public safety.
“You can do that to just make people safer.”
Listen at 41:28
A data-center moratorium provides time to design broadly beneficial incentives.
“the point of any moratorium is a pause while you are figuring out the broader incentives that make it work for everyone.”
Listen at 47:19
Data centers create genuine energy-consumption and pollution concerns.
“The energy consumption is real and the potential pollution is real.”
Listen at 47:58
Data centers commonly seek speed by using natural-gas turbines instead of the grid.
“The default right now where companies are trying to do it as fast as possible is that the fastest way to do it is not hook up to the grid and just use natural gas turbines”
Listen at 48:53
Data-center demand can be used to incentivize beneficial infrastructure investment.
“we can set up the incentives to use that to our advantage.”
Listen at 51:48
Data centers should pay all costs plus fees to communities, grids, or states.
“you need in any of these developments for them to not just cover all of their costs not just bring all of their new energy but also pay a fee on top of that to the community, to the grid, to the state”
Listen at 55:53
Data centers should pay more than their construction costs.
“they should be paying above and beyond just their costs of building.”
Listen at 56:23
AI development is risky, but meaningful intervention remains possible.
“I think we're on a very risky path, but one that we very much still have the time to change.”
Listen at 56:48
AI governance can achieve adequate control through action.
“I am an optimist. I think we can get there in AI.”
Listen at 57:20
Illinois SB 315 is the only U.S. bill requiring audits of AI companies.
“SB 315, which is the only bill in the country that requires audits of these companies.”
Listen at 58:40
AI’s power is a reason to impose regulatory guardrails.
“because it could be so powerful is the reason why you want to put some guardrails here.”
Listen at 1:01:37
AI deployment requires strict regulations and effective public protection.
“you better have actual strict regulations and you better actually be protecting people”
Listen at 1:01:52
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.