Loading summary
Transcript99 lines
- [00:00]
Chase Sapphire Advertiser
This year's girls trip to Telluride was the best. We one upped ourselves with my Sapphire Preferred card. And with 5 times points on Chase Travel, plus 3 times points on vacation homes with top brands, we got this incredible cabin.
- [00:12]
Molly Roberts
It was a mansion.
- [00:13]
Chase Sapphire Advertiser
And with three times the points on dining, we ordered a wagyu steak dinner, and that pistachio gelato was too good. So where should we go next year?
- [00:21]
Scott R. Anderson
I've got ideas.
- [00:22]
Chase Sapphire Advertiser
Chase, Sapphire preferred the card that's preferred for a reason. Cards issued by JPMorgan Chase bank and a member FDIC subject to credit approval terms apply.
- [00:33]
Scott R. Anderson
Kevin, you got to do a little. A little World Cupping recently. Is that right?
- [00:37]
Kevin Frazier
Did. Did some World Cupping. So I hung out with my sister who lives in Seattle, and we went and saw the Belgium vs Egypt match. And the World cup is a battle for the ages. A battle for the ages. It was, you know, quite, quite the scene. It was difficult because Belgium and Egypt both wear red, and so all of the fans. It was impressive because everyone was wearing red, but we just didn't know who was who for.
- [01:06]
Scott R. Anderson
So you don't know who to punch when you lose?
- [01:07]
Kevin Frazier
Yeah. Right. So, thankfully, the fighting didn't break out because no one knew who was on their side or not. But Seattle knows how to host a World Cup. It was a hoot and a half, as the kids say.
- [01:18]
Molly Roberts
Who were you rooting for?
- [01:20]
Kevin Frazier
So I flipped a coin and therefore was for Egypt. My wife was therefore in favor of Belgium. And so I think we. We ended up with a tie. So everyone went home happy to some extent.
- [01:33]
Molly Roberts
I joined a World cup sweepstakes thing where you're randomly assigned the team that you get. So it's supposed to be exciting because maybe you don't really have anyone you're rooting for, and then you get another team to pay attention to. But I got Cutter. There's just no point.
- [01:50]
Kevin Frazier
They're already out. I believe they've lost a chance. Well, Molly, you gave it your best shot. You really tried with Team Cutter.
- [01:57]
Molly Roberts
I did.
- [01:57]
Roger Parloff
I did.
- [01:58]
Molly Roberts
I was really rooting for them. I didn't give them any strategic advice or anything, but that might have been to their advantage.
- [02:03]
Scott R. Anderson
I'm not gonna lie. Everyone in the Persian Gulf needs kind of all the little silver linings they can get. So it's kind of a bummer, but you're throwing. You're throwing them so much needed help at this particular moment.
- [02:13]
Kevin Frazier
Roger, are you leaning into France? And Mbappe, he's destroying the competition. He's had a great World cup liberte
- [02:21]
Roger Parloff
Fraternite and Mbappe
- [02:26]
Scott R. Anderson
Foreign. And welcome back to Rational Security, the show where we invite you to join members of the lawfare team as we try to make sense of the week's biggest national security news stories, whether they are in our lane or not. I am your host, Scott R. Andersen. Thrilled to be back here with you for another week with a wonderful all star panel of my colleagues. Joining us this week is a trio of lawfare senior editors. Rarely do I have my job lined up so easily for me and describing people's titles and affiliate. But joining us first is lawfare senior editor as well as an AI fellow at, I should say UT Law School in Austin, Kevin Frazier. Kevin, thank you for joining us.
- [03:07]
Kevin Frazier
Always a pleasure, Scott. Thanks for having me.
- [03:09]
Scott R. Anderson
Also joining us is lawfare senior editor Molly Roberts back on the podcast. Molly, thank you for joining us.
- [03:14]
Molly Roberts
Thank you.
- [03:15]
Scott R. Anderson
I recently assigned Tyler a masked assignment for how regularly I impose upon him to come on the podcast. And you are coming up a short second. I don't know if you're fully masked yet, but you're probably approaching jib status, which I think is pretty good. So I would take it. That's a good one. I think I could win.
- [03:30]
Molly Roberts
My odds are better than cutters.
- [03:31]
Scott R. Anderson
Yeah, there you go.
- [03:32]
Roger Parloff
Exactly.
- [03:32]
Scott R. Anderson
Exactly. And joining us for our third panelist is none other than law firm senior editor Roger Parloff. Roger, thank you for coming back on the podcast.
- [03:40]
Roger Parloff
Oh, great to be back, Scott.
- [03:42]
Scott R. Anderson
Wonderful. We have a real set of interesting stories, all in the zone of AI or AI adjacent legal actions, all of which kind of lined up and all of which are worth talking about with a little bit of politics thrown in for good measure. So let's get into it. Our first topic this week, citizen Kaint can't Kaint when the NAACP sued Elon Musk's X AI under the Clean Air act, alleging that the company built dozens of gas fired turbines to power a data center in Mississippi without relevant air permits and exposing nearby predominantly black communities to harmful pollution. The Justice Department has opted to do something it has never done before. It has intervened in that citizen suit against that private company in order to kill it. DOJ's motion offers two theories. First, that shutting down the turbines would threaten national security because the military relies on xai's GROK model, including in relation to the Iran war, to secure the nation. And also because the Constitution's vesting of executive power in the president means private citizens cannot enforce federal law over the executive's objection, or at least this federal law over the executive's objection I should note how strong are these arguments and what would it mean for environmental and other citizen enforcement lawsuits if DOJ were to prevail? Topic to Crock the Vote we may be living through the first true AI elections. In Manhattan's 12th district Democratic primary, more than $40 million in AI industry and AI safety money was spent trying primarily at least on the part of the AI industry to defeat a little known previously Assemblyman Alex Boris, in what rapidly became a something of a national referendum on whether voters care about AI regulation and AI safety bores, ultimately lost to arrival, although notably that rival, also a supporter of AI safety legislation. So a little bit perhaps more of a mixed picture than some would argue. Meanwhile, overseas in Malaysia, parties are using AI do multilingual voice notes to use chatbots and other new AI driven technologies to reach out to voters in new and novel ways, many of which we are likely to begin seeing if we haven't already started seeing them here on our shores. And just this week in Washington, a new study has concluded that Frontier AI is perhaps more persuasive than ever, but also may not be as politically neutral as some suspect or one might hope. What does this all mean to democratic politics when both the money and the messaging involved in our politics are increasingly shaped by AI and topic 3 kill kill switch. Kill kill. The government's frontier AI kill switch is now ready to have its first day in court. If you recall, a few weeks ago, the Commerce Department's Bureau of Industry and Security sent Anthropic and as informed letter ordering it to suspend all access to its Fable 5 and Mythos 5 models for any foreign nationals, including those within the company, including its own employees. This ultimately led Anthrop to pull access to those models for everyone within a few hours. But this past Monday, a technology startup called Legion Legal Tech filed a lawsuit against the US Government alleging that it has acted in a way that is unlawful, inconsistent with relevant statues, and raises a number of statutory and constitutional concerns. How strong is the legal challenge and what does it tell us about whether courts rather than the executive, will end up defining the government's power to switch a Frontier model on or off? So for our first topic, Roger, I want to come to you. We're seeing a pretty novel legal action and kind of in some ways actually what is, I would say, a fairly conventional legal action next to a fairly novel legal action. The NAACP filed a lawsuit against XAI under a citizen suit provision that essentially allows, I think, anyone, I don't think really limited anyone affected by air pollution of a certain type done in violation of the CIA, at least allegedly in violation of the CIA, to seek both injunctive relief and then to also seek monetary damage. Pretty substantial monetary damages against alleged violators, but those damages get paid to the US treasury, not to those individuals in question. Notably, this statutory regime employed by the NAACP is now being invoked in a unique ways, both by the US Government, but notably, I should say, also by Xai. The US Government's brief basically came in and supplemented an argument made, I think, on the same day, filed on the same day in the motion dismiss filed by Xai and the other defendant in the litigation, where they essentially argued, Xai argued, hey, look, as set up, the citizen suit provision is unconstitutional because it infringes on the authority and the constitutional authority of the President to take care that the lobby faithfully executed the enforcement power of the President and the executive branch, and that you can't give citizen suits of this type to this model, particularly where they're giving money to the treasury. That was a hook for them. And now we see the Justice Department intervening, making the same argument, essentially saying, look, in particular, they focus particularly on the fact that these damages are supposed to accrue back to the executive branch. They're saying, you can't give a citizen the ability to pursue enforcement action on behalf of the executive branch. That's not what we're allowed to do. We have the sole authority. And they invoke the False Claims act and KITAM action, a type of action the False Claims act is our most prominent example of, and some case law around that, to argue that's why we have this ability to intervene and essentially quash this claim by these plaintiffs. Talk to us about how this fits into the kind of broader controversy of this case, of this issue. And I should note, also, they came in, the government came in with, I think, the much more conventional part of the brief asserting, essentially, look, granting this lawsuit, which, if the plaintiffs got what they asked for, would essentially mean shutting down these data centers, or at least the way that they're powering these data centers, would be bad for US national security because it would impinge the development and use of grok, a model that the US Government relies on that's less conventional. It's not weird to see a statement of interest by the US Government saying, here are the policy concerns to be implicated, whether we agree with them or not. But it's interesting context to see that aligned against this other argument. What does this stand out for you as, like, how weird is this and what do you make of these different arguments? How credible do you find them?
- [09:37]
Roger Parloff
I think the comparison to the Ki Tam suit is not very strong. Ki Tam suits are brought. Really the entire suit is brought on behalf of the government. These are not brought on behalf of the government. The statute creates a separate cause of action for citizens, and it creates it in case the government doesn't pursue it. For 50 years it's been used that way. And this is the first time the government has invoked, has said that they have a right to end citizen suits. There is a provision in the law that says if the government is already enforcing the law against pollution, I mean, the goal is to eliminate pollution. It's not just to get damages to anybody, it's to stop the pollution. In the past, if the government was in fact already trying to stop the pollution, then you could say, yes, we don't need a citizen lawsuit on top of it. But it's never before been the case that you could say the government doesn't want to stop the pollution and therefore the citizen can't try either. And that's what this does. It would also, obviously, it's not just the Clean Air act, it's all of the environmental laws. I mean, not all, but it's many. The Clean Water act and RECRA and some others. And in fact, if you accept Xai's argument, which is even broader, which is that citizen suits themselves are unconstitutional per se, then you get even further out. And we have civil rights laws that individuals are allowed to bring in addition to the government. Those would be endangered. So it's an extremely. The impact would be enormous if this were accepted in its strongest form.
- [11:46]
Scott R. Anderson
So I want to drill into some of the legal technicalities a little bit more. I've done a little bit of work in key team actions, and I agree with you, Roger. I'm not sure the government or X AI are frankly quite on good footing and trying to draw such a tight parallel. And even if they did, I'm not sure quite lead to the outcome they want. Before we do that, though, let's take a step back and think and put this lawsuit in the context of the burgeoning AI industry. And Kevin, I want to come to you for that. I mean, talk to us about this point. We know AI development and use is energy hungry. It takes a lot of energy to do it, and that has environmental effects. But and that we're seeing this particularly in those communities near these big data centers that become clustered in different parts around the country. Those are becoming increasingly controversial and there's been lots of reporting, particularly in the last few months, I feel like, about communities pushing back against efforts to locate data centers close to them because of concerns about sometimes environmental effects, sometimes the economic effects that come with them, the other sort of effects. Talk to us about how this conversation is fitting into the narrative around AI law and policy. Like, are these lawsuits? Is this a widespread view within the AI community that these lawsuits are a fundamental threat? Is XAI maybe a little bit ahead of the game or perhaps a bit of an outlier and willing to make such as Roger said, like pretty broad arguments. Again, if anything, DOJ is a narrower version of this than what XAI actually advances in its mission to dismiss about the limits here. Where does this fit into that broader landscape?
- [13:16]
Kevin Frazier
Yeah, to quote a fantastic movie, I would just say it's a mess out there in terms of trying to make sense of all of the data center politics. We know that from polling. It's still not the case that the majority of Americans, let's say, have a blanket opposition to data centers. There is a strong and I'd say growing number of Americans who are concerned about data center development unchecked. And I think that this narrative building around this suit in particular, only adds fuel to the fire of folks feeling like they have been shut out of the process of having a voice through their local government as to how and when these data centers are going to be developed in their communities. We know that in particular there have been allegations that xai, for example, has not been as transparent as other labs, let's say, or as other data center developers in how they are working with communities and what sorts of benefits and or extractive efforts will be involved with that data center development. So to see that XAI is the target of this suit, and then to see furthermore that they're advancing a approach that would limit the ability of citizens to contest data center development that is certainly not going to increase or improve the public perception of this data center build out. So I'll just flag that as as one issue. The second issue is going to be around what are the true costs of data centers to our energy grid, to land use and to water. Shameless self promotion for a different Lawfare Podcast we have a excellent podcast coming out with Andy Masley on scaling laws that dives deep into the weeds of water usage and land use and electricity rates. But this is the crux of the issue is to what extent are we seeing that data centers are actually exacerbating the climate concerns that folks may have or the electricity cost concerns that folks may have. And when it comes to looking across the different range of data center developers, XAI traditionally has stood out as perhaps being more willing to use less clean sources of power for their data centers, which makes them a bit of an outlier on some of these data center conversations. One of my very tired phrases of many is today's data centers are the worst data centers we will ever use. It's in the interest of Microsoft, of OpenAI, of anthropic, to be more efficient and to find better approaches to powering and running their data centers. But as of now, the technology that many of the labs are relying upon, including xai, aren't being run as cleanly, I think, as many communities would prefer.
- [16:04]
Molly Roberts
Yeah, I found what the Department of Justice did in the case pretty extraordinary. I mean, it seems like a really general and radical thing to suggest that the government, if it doesn't want to enforce, can also null displace private enforcement. It's not just if it is enforcing, it's if it's not enforcing too, which seems to kind of take a lot of power away from Congress if there's this citizen suit provision and the government says, well, no, not if we don't want the citizens to sue. And of course XAI is arguing that citizen suit provisions are themselves unconstitutional. And then when it comes to the policy implications, not just for environmental laws generally, which obviously there are huge policy implications, and not just for all those other statutes like whistleblower, civil rights, various consumer safety statutes that also might have citizen supervisions, but specifically for AI, it seems pretty dramatic because if we're talking about stuff like the establishment of data centers at the moment, unless Congress legislates on them, this is really the only constraint and it would be doing away with it if the government is going to be consistent here. So I think the implications are pretty big. Congress could legislate on data centers in some form, or it could legislate on the citizen suit question. It could say that the government can't come in and seek dismissal, but I wouldn't necessarily expect it to do that because it doesn't do all that much.
- [17:43]
Scott R. Anderson
So let's dig into the legal technicality of this a little bit. Roger and Kevin and Molly as well, if you guys have thoughts on this. So they are trying. The government really leans heavily on some case law in the Ketam and False Claims act context, which strikes me as a little bit of not quite the right parallel to draw a Ketam action traditionally is a case where the government has an interest and it authorizes citizens to enforce the government's interest on its behalf by initiating a lawsuit. This was actually super, super common practice early in the Republic. It has now become less common. There are actually lots of Ketam statutes still on the books. Like I did account for a paper I wrote a couple years ago. I think there's something like two dozen, maybe more. But the only one that we really use anymore, in part because the others have been narrowed substantially by the courts, is the False Claims Act. The False Claims act does allow the government to essentially decide, okay, we are going to take over a lawsu and then potentially settle it on its own terms, like do away with it if it chooses to do so. There are some process and restrictions around that, and I'm not going to pretend to be a full expert in it. But there is that element of executive control and there is some case law and a lot of academic writing that strongly suggests that's how you reconcile this particular the False Claims act with the generally viewed exclusive authority of the executive branch to be involved in law enforcement. Although again, I don't believe there's a ruling squarely on this point suggesting that it's absolutely necessary in this context. I don't think that that that's been as firmly decided. Although perhaps I've missed something in the last few years since I've looked at this. But here, this particular provision, it's really telling to look at the actual text. I think this is 42 USC 7604 citizen suits in the Clearing Air Act. It says right off the front, the authority to bring civil action says any person may commence of civil action on his own behalf. So you, you, this is not a Ketam case where you don't. The reason Kitam cases are controversial is because individuals who wouldn't otherwise have standing because they couldn't articulate an injury as a result of action are able to pursue a lawsuit and secure damages to them. That's part of the False Claims act regime, is that they actually get a slice of whatever damages they're able to claim on behalf of the government to motivate them to pursue the lawsuit they're suing on behalf of the government. Here that's like expressly not the case. And the naacp. Part of the argument that Xai advances is that they don't have standing because they can't sue on behalf of their own members in this context. I don't think that's actually a very persuasive argument in this context. But like they clearly are making a coerced argument. They're not asserting that they're suing on behalf of the United States. They're very clearly claiming personal injury. The only mechanism that ties us back to the government is the fact that civil penalties that the court is authorized to impose go back to the federal treasury. Although notably actually in this provision, they don't exclusively have to. There are situations where penalties can be spent and the court's discretion on beneficial mitigation projects which are consistent with this chapter and enhance the public health or the environment up to a certain amount. Although I think the amount of controversy here is maybe over that threshold. So am I wrong on this?
- [20:45]
Kevin Frazier
Like why?
- [20:46]
Scott R. Anderson
How persuasive? This is a key tm. Can you draw these sorts of parallels? Isn't there just a fundamental problem here to say that this is in any way suing on behalf of the government when it's not? These people have to sue on their own behalf and then it limits the remedies. And notably with the penalty is one remedy. The other remedy that they're asking for in this case is injunctive release, and that would be injunctive relief in response to a claim of somebody trying to vindicate their own interests and legal rights. So why is that an obstacle here? How does the DOJ try to work around that?
- [21:14]
Roger Parloff
Yes, I agree with all of that. We haven't yet gotten the NAACP's brief, but that is obviously the main response. And I think just talking a little about the facts also would be useful because a lot is not disputed here, including the fact that Xai did not seek a federal permit. I mean, you know, you can build the data center if you seek the federal permit and show that you're using the best available control technology. Why not do that? The name of it is Colossus too, which should give you some idea of where we're going.
- [22:00]
Scott R. Anderson
These AI companies and their marketing, I have to say they really should do themselves some harm in emphasizing their own enormity and significance.
- [22:08]
Roger Parloff
Yeah, and you could tell this is an Elon Musk operation. And it had 27 gas fired turbines when the suit was filed. They're up to at least 57 now. Each one is apparently the size of a single family residence. Again, it's not disputed that I don't believe that they spew nitrogen oxides, that they spew carbon monoxide. The fine particulate matters. I don't think it's disputed that those cause are associated with heart disease, lung disease, other serious illness, premature death. And it's not disputed that they didn't Seek the permits. And so the NAACP comes in on behalf of the neighborhood. It's a poor neighborhood, apparently. It's really sort of greater Memphis, although it's located in northern Mississippi, on behalf of a heavily black community. So they're suing on their behalf, on the behalf of their children. They want to breathe clean air. And this is true of all of the citizen suit provisions of pollution. It's about people protecting their own interests in clean air. So I think you're exactly right. The Kitami argument is a weak analogy.
- [23:35]
Scott R. Anderson
I have a theory about what's going on here. Actually, Kevin, I want to come to you about this and tell you something that makes sense from a perspective. So they have colorable arguments here as to why they may have complied with this. Notably, state of Mississippi has a state implementation plan, sip, that's required under the Clean Air Act. They've enacted it according to Xai. I haven't seen this contested, although maybe I missed it somewhere in the record. The state of Mississippi said you don't need to take an additional measures. They've kind of more or less signed off on this. They don't see this as an issue under the Clean Air act under their own law. And so, and the EPA doesn't appear to be second guessing this under this administration lawsuit which has approved the Mississippi sip. So you've got these arguments that are still in the mix saying, well, look, we complied with the relevant state law. The agency doesn't deal with that. Now I think colorably, that's the reason you have these citizen supervision, so that they can subject these determination the federal government, the state government to judicial review and see if they're actually compliant with the statute. But regardless, maybe Xai is right, they have these statutory arguments. But one thing you get when you raise these novel constitutional arguments is time, because you have to litigate them. This court's going to be pretty comfortable on the factual record and what it understands about the statute, assessing that kind of basic administrative law, statutory interpretation sort of argument. And yeah, you're going to have to appeal that and go that up and down. When you throw a lot of these other constitutional arguments into the mix, you're creating a lot of grounds for multiple rounds of appeal. And that's not always a. It's really bad for people on the ground. There's a reason to object to that. And you worry about preliminary relief in the interim. If you were the defendant of these cases, you worry, well, if we drag this out too much, they may just give them preliminary relief and therefore we're going to be in a position where we're going to pay a bunch of money even if we win in the end. But here you've got the US Government saying for national security reasons, we can't shut these things down. I think particularly this court has suggested those sorts of foreign relations, national security arguments weigh really heavily in the preliminary relief sort of scale and suggest as any preliminary relief is not going to shut down these reactors, maybe harm will, you know, penalties will crew things like that, but they're not going to shut down these reactors. And now you've set up multiple issues that will have to be litigated or at least could be litigated up to the appellate courts, maybe the Supreme Court potentially, although I'm actually not sure this argument is colorable enough, the Supreme Court will actually feel the need to take it up. And that extends the potential lifespan of this litigation. I'm wondering, could this be a deliberate action on xai's part, a little legal brinksmanship where when you're involved in these super, super competitive races between AI companies and they're desperately trying to throw everything they can in developing models, utilizing models, making them available, perfecting them, that they just being able to kick the can down the road may bear dividends for them, even if it proves kind of expensive. That's kind of what I think might be happening here. But I'm curious about your response to that, Kev, whether you think that's feasible or not.
- [26:27]
Kevin Frazier
I certainly think it's feasible. When it comes to the reliance on Colossus 2 and related data centers of that scale, it's huge. I mean, when you go and you read anything about what the hyperscalers are looking for at this moment, they just need more and more data centers to be built. And so far we've seen that in counties and communities across the country, there have been efforts to stall those projects or delay those projects, and in many cases they've been successful. And so the labs are becoming a little bit desperate to find any compute that they're able to access so that they can train the next model and so that these enterprise users can continue to deploy these models in highly sophisticated and highly compute intensive fashions. And so for something like Colossus to go offline would be something that not only hit the bottom line of xai, but would impact the entire AI ecosystem as we continue to be behind the curve of data center development with respect to what's needed by the industry. And Scott, to your point, in terms of this national security argument, it's not going to be lost on the court. And I'm sure we're going to see this be raised in even more detail. The fact that data centers were targeted in the war in Iran, in Yemen, I believe was the first data center that was targeted. And so in terms of making a national security argument, it's not going to be too difficult. And I do think that will end up being pretty dang persuasive to the court in this context. Now, now what's unfortunate, and to just riff off of what Molly was saying here too, is this is another instance at which AI is really pressure testing our constitutional order and pressure testing separation of powers. Because here we're seeing that the executive is just able to move so much faster than Congress and the courts are not going to. And I'm not advocating for this. The courts aren't all of a sudden going to go into turbo mode and adjudicate all of these AI claims faster than everything else just because, you know, Elon might like that or not like that in this context. So we're seeing that Congress is a little flat footed, the judiciary is acting as intended. But if the executive can move so much faster than the other branches and act in so many myriad contexts such that they, you know, change the kind of political conversation and legal conversation, well, that's not the system operating as intended. And just to finally end my rant, I've been able to talk with some of the folks in Memphis on the ground who are fighting these battles and who are talking with affected community members. And it's just so unfortunate that we couldn't have had more local engagement from the outset and just more transparency from the outset so that we could have seen data center development bring communities along with them so that everyone would have benefited from not having this sort of, of more adversarial approach and one that's going to just waste a bunch of money on lawyers fees, which as a lawyer I don't even condone.
- [29:28]
Scott R. Anderson
Yeah, I mean like the pressures of this and this administration's focus on them is really notable. And I'm not sure it's irrational. To be clear, I'm not 100% critical of this, although I do think we have to be cognizant of environmental and other costs we seen as administration takes some pretty dramatic steps towards things, even things like deploying small traditionally strategic nuclear reactors potentially being used to help help fuel data centers, other sort of sites. I don't know if that full step has been taken there, but we've seen steps in that Direction. It's been talked about a lot. And this does kind of fit into that. What does strike me as like kind of extraordinary though, is the fact that you see the Justice Department buying into this legal argument that I think is bumpkus. I like don't think it's a strong legal argument. Right. Maybe I'm wrong. I think maybe they could get away with. Maybe you can impose fees and penalties. Right. But when you're talking about the injunctive relief, somebody's suing for the benefit of their own harm. It's really hard to see how this is something that US Government has a claim over just because any fines happen to accrue to the Treasury Department. So it's a pretty bold statement. And the fact that you see the government. I'm less surprised by the statement of interest. I 100% would have expected that, honestly, from this administration in this case. But the fact that they're pairing it with this argument, and particularly if I'm right, that it is an example of industry using a questionable legal argument to extend the timeframe without really hinging on in a way that doesn't hinge on the merit of that argument. That's a troubling thing to see the executive branch of the Justice Department play along with.
- [30:51]
Molly Roberts
With.
- [30:51]
Scott R. Anderson
I have to see, and I will say Stanley Woodward has gotten the most press attention because I think he's one who actually signs this. But it's signed by a number of DOJ officials, I believe all political appointees, but up and down, including enrd, the Environmental Natural Resources Division of the Justice Department. So it's got buy in from a variety of people putting their name on this. And again, I'm not sure it's. We'll see what happens with additional brief, additional argument. I'm not persuaded that it's a particularly credible argument, at least when it comes to the injunctive relief. And it's a little troubling to the Justice Department to take that position.
- [31:20]
Roger Parloff
I think I'd just like to add there's another noxious, potentially noxious aspect to this, which is a corruption opportunity. We've talked a lot about abusive actions by DOJ that are vindictive against enemies of the President. Another type of abusive potential government action is those that are beneficial to friends of the President. And maybe coincidentally, this data center is run by Elon Musk, who this week at least is a friend of the President. And so you could see if the President has the power to turn off pollution suits brought against major industries all over the country. Water pollution Clean air, other sorts, just by the snap of the fingers, it's a huge incentive to give donations to that president and try to persuade him to get you out of a jam. So that's another huge factor that if this theory is accepted, you've given another weapon, another pressure point into the hands of the executive.
- [32:39]
Scott R. Anderson
Yeah, I think that's a really, really good closing point, Roger. And again, these citizens provisions, I mean, in theory, if you took it too broadly, while they really tie this to this provision and in their arguments to saying, oh, this is because at least in these cases where the funds go back to the treasury, but other statutes might have similar regimes. I don't know about the Clean Water act and other statutes like that. But more fundamentally, they're not actually clearly limiting their argument to those cases. It's potentially it could go to other citizen suit arguments, which is what Xai is arguing. So that itself would be much broader ramifications, basically not only limiting the executive branch, but preventing Congress's ability to extend personal liability in some cases. The Justice Department doesn't go that far to their credit yet. They are really focused on the fact that these fees go back to the Treasury Department in their brief, but they also are extending it to injunctive relief in a way that doesn't isn't tied to those fees. So I'm not sure we can put entirely beyond it a broader expansion of this theory in future cases. So let us now turn our eyes from the legal realm to the political realm, because we have had an election day this past Tuesday. Frankly, we've had like lots of election days recently. This is election day season, particularly if you pay attention to Primari. We had another tranche of primaries in New York and a number of other states, but I've mostly been seeing New York results come through this past Tuesday and some overseas elections that have been kind of notable in Malaysia, where we're seeing a lot of novel uses of AI, even as in the United States. I'm sure there are some of those uses, although I haven't seen them, but I'm sure they're baked into various political ads and promotions I've been seeing. But then in particular, we also are seeing the issue of AIs, particularly safety around AIs, to some extent. This data center and the cost of data center issues as well, increasingly playing a role in US Politics. And then money from AI, the huge amount of money that AI is generating, playing a part with large sums going into small races. In the case of the Alex Borres case in New York, which is, I think, New York's 12th congressional district, really out some $40 million in a race that size. Pretty insane, driven by AI money on either side of the safety debate. And it intersects. Kevin, you brought attention to our attention as we're talking about what to talk about for this episode. It intersects with some really interesting scholarship we've come out about the persuasiveness of AI, its political bias. So talk to us a little bit about. About what some of the scholarship is pointing towards indicating about how AI can be used, its effectiveness and its impartiality potentially, or lack thereof, in these sorts of contexts, and how it begins to intersect with these new uses and relevances of AI we're seeing in the electoral realm.
- [35:05]
Kevin Frazier
Yeah, and I think just to build off of your point earlier, AI is certainly on the ballot in many different contexts. Whether it's data centers, as you noted, whether it's potential job displacement, whether it is use in schools. All of these questions are issues that candidates are increasingly talking about and running on. We actually saw that the Senate president in Utah lost his primary on the question of data centers, because in Utah there's been a particularly contentious fight around a data center proposed by none other than shark Kevin o'. Leary. And that's been subject to a lot of controversy. And so we know that AI is an incredible issue when it comes to determining how some of these races are going to play out. But I want to focus in particular on how AI is going to change the nature of democracy and the nature of our kind of electioneering to begin with. And this is really building on research that came out from Hackenberg et al. Which is AI systems out persuade expert humans and to just nerd out for a second, because I think this methodology is. Is important to flag and interesting for listeners. Imagine you get a number of issues presented to you, and you need to place your stance on a certain controversial topic. From 0 to 100, you enter a score of 71. Then you are randomly assigned to either chat to a human or chat with an AI for about 14 minutes about that topic. And then at the end of that exchange, you then go again and from 0 to 100 rank, where you land on that issue by virtue of talking to either the human or to the AI. And this study tried to measure did your stance change on that issue by virtue of talking to the AI or talking to the human? And when I say talking to humans, these researchers recruited the best of the best in terms of human persuaders. So we have elite debate champions who were Recruited. We have folks who were trained using AI to be more persuasive and to try to be all the more compelling. And then we had incredible incentives that were given to these persuaders. So some of them would receive a bonus of, for example, £250 if they were the most persuasive of the humans. And so we have humans, you know, on steroids of persuasion here, really trying their best to be as persuasive as possible. And when they were paired against AI, they lost. And they lost in a dramatic fashion. And we saw that when it comes to the persuasive power of AI, it is a order of magnitude more persuasive than the next best human. And that has huge ramifications on our election processes, because we know, as you flag Scott, in places like Malaysia, political parties are actively asking, how can we integrate AI into voter outreach, into message design, into just basic organizing? And if we know that AI is vastly more persuasive than the next best human, well, that's a wildly different kind of in election that we're seeing go on. And we really have to start to ask questions about the agency of voters, about the autonomy of voters and their ability to make sure that they're actually acting on their values. Values, and not the values advanced by whatever AI system they're engaging with. Now, critically, researchers did find that if you constrained the AI in two specific ways, then you kind of defanged the persuasive power of those AI tools. So the first was to limit how quickly the AI could type. So if you didn't see that AI was being rapidly responding, whereas that elite debater was clacking away trying to get their responses, and that definitely had a limiting impact. The second thing, and this is just hilarious to me, is if you reduced the amount of content the AI could spew at the recipient, that also significantly limited its persuasive effect. And so you could actually make AI on par with humans by virtue of just lowering the speed and limiting the amount of facts that it threw out at a voter. So. So this has incredible importance to me as a born and raised Oregonian. We are the home of the initiative and referendum. And when you begin to think about how, let's say a sponsor of a ballot initiative could deploy a persuasive AI tool and suddenly find and galvanize a lot of support for a really bad ballot initiative, we have to ask, what safeguards are we going to impose such that we don't see entire elections swayed just by virtue of one party or one sponsor having access to the best AI, whereas everyone else is still clacking away at the speed of an expert debater.
- [40:14]
Scott R. Anderson
Hey folks, Scott R. Anderson here. You know, summer has arrived here in Washington D.C. and it completely rewires how I dress. Things may still seem cool and comfortable now, but by mid June it's swampy out. I'm hauling the kids around town on my bike half the time and working the grill in the backyard the other half. I need clothes that are light, breathable and still look like I made something of an effort. And that's why I keep coming back to Quince. Their Pima cotton pocket tees have basically become my uniform. They're soft, they hold their shape, and they don't look like an undershirt on hikes and bike rides. I turn to the Flowknit performance tee. It moves and it doesn't hold the heat when I'm cooking outside or out on a casual dinner. It's the organic stretch Geno shorts comfortable enough to crouch down in the garden in, but nice enough to stop by that place around the corner afterwards. And the 100% organic cotton short sleeve oxford is my go to when dad at a backyard dinner needs to read as a notch more put together. And here's the thing. Everything at Quint's is priced 50 to 80% less than similar brands. They work directly with ethical factories and cut out the middleman. So you're paying for quality, not a brand markup. It goes way beyond clothes too. Quint has cookware, bedding, even sofas. It's the kind of place you end up recommending to everyone. So elevate your summer wardrobe. Go to quint.comsecurity for free shipping on your order and 365 day returns. Now available in Canada too. That's Q-U-I-N-C-E.comsecurity for free shipping and 365 day returns. Quint.com Security now let's get back to the show. For investors at the highest level of sophistication, those with strong advisory support, the natural outcome is two distinct portfolios, one public, one private. Each plays a different role in the overall strategy. Uncover new investment opportunities for you@creativeplanning.com New
- [42:03]
Livy Dunn
I'm Livy Dunn, All American gymnast and Vuori athlete. When you travel and train as much as I do, you find happiness where you are on the mat or on the sand. Movement and comfort are essential. That's why I live in performance joggers by Viori, made from Dream Net fabric that's made of 89% recycled material, effortlessly soft and made to move as much as I do. My happiness starts here in the softest joggers on the planet. Get 20% off your first purchase at Vuori.com Libby that's V U-O-R-I.com L I V-V-Y exclusions apply. Not only will you receive 20 off your first purchase, but enjoy free shipping on US orders over 75 and free returns. Go to Vuori.com Libby and discover the full versatility of Vuori clothing exclusions apply. Visit the website for full terms and conditions.
- [42:56]
Scott R. Anderson
Yeah, it's a really interesting study. I want to spend more time on it. I read it in preparation for this episode. I did not get to deeply read it on methodology. It reminds me of anybody who's done debate has seen particularly parliamentary style debate. I never actually did debate. I did a lot of debate adjacent stuff. I think it's parliamentary style debate where you'll see the world champions rattling off as many facts and points as possible, in part because they are gaming the system, because part of the metric that traditionally is used is checkboxing on number of issues raised and number of addressed. And it kind of strikes me as that it's so interesting to think about how we persuade authority and this would prove more persuasive in an era when expertise is under so much doubt and there's so much lack of trust in institutions because in some extent you would traditionally expect the person able to most mimic this methodology of deploying facts and styles would be a genuine subject matter expert. And yet the trust in those, the sense of credibility has been on the decline for a number of years now, in part because of broad lack of social trust. So I have to wonder whether I want to look into exactly the more the context of which these were deployed and the types of what we're thinking of as persuasion. Because if it really is a matter of volume and speed, that's interesting. And you can see cases where that would make a big difference. But you can also think a lot of frankly the most important cases like politics, where it's actually not, or it's at least not universally important.
- [44:21]
Kevin Frazier
These were matters of UK political debate that they were tested on. Yeah.
- [44:27]
Scott R. Anderson
So it was issues based as opposed to like a voter perspective, correct?
- [44:30]
Kevin Frazier
Yes. Yeah. Issues based in the uk.
- [44:32]
Roger Parloff
Yeah.
- [44:33]
Scott R. Anderson
Yeah. That's interesting. Roger.
- [44:35]
Roger Parloff
Yeah. I just had a question for Kevin and I guess you might have answered it by the end, in effect. But I gather these are all written arguments. It's not oral arguments.
- [44:47]
Kevin Frazier
Yeah, so this was as if you were engaging in, like, a text exchange. Change with something. Okay, yeah, so we'll see. You know, we're. We're also seeing that AI tools are becoming more multimodal. There was an instance during the super bowl year of elections in. In 2024, where the candidate for president in South Korea was using AI avatars such that the AI avatar would be deployed in novel situations and made more funny. For example, in if the candidate was supposed to be engaging with a younger audience. For example, the AI avatar was released in a way that he'd be slightly less boring. Apparently, he wasn't a very energetic candidate. And so we saw in that super bowl year of elections that the sky didn't fall. Some people thought that was going to be the election year in which AI suddenly changed everything. And 2026 is on the horizon. And we, again, haven't seen AI change. Change everything yet in these primaries. I will note that here in Texas, we've had debates around candidates who have been releasing ads that have used AI. For example, there was one attack ad on Talarico where he was reading his tweets. This wasn't actually Talarico reading his tweets, but it was an AI version of Talarico doing so. Is that okay? Is that not okay? Today we're talking. On June 25, we saw a bipartisan bill drop in the Senate, which is the AI Label Labels act, that would require more disclosure of when and how an AI generated image is being shared. So there's just a lot for us as a political community to figure out about our level of comfort with AI in these sort of electoral processes. Some people you talk to will say, oh, well, we've had yellow journalism since, you know, 1901, and we've had political cartoons since even before that. This is just a new caricature or a new attempt to kind of change the disco course. Based off of this recent research, I think it's hard to make that sort of argument given just how persuasive AI can be.
- [46:54]
Scott R. Anderson
I may want to make it. Anyway. Before I do, let me. Let me bring in Molly on this. Molly, you know, I know you are somebody who is, you know, thinks about the political context of these sorts of things and looks at our political discourse and how it intersects with a lot of the policy and legal issues we look about. What do you. What jumps out to you about this? What do you find concerning. What do you find interesting?
- [47:12]
Molly Roberts
Yeah, when I think about AI in elections, I guess I think of it showing up in two ways, both of which Kevin has touched on. So the first one would be when it comes to deceptive content, whether that's something like a deep fake or shallow fake or whatever of some candidate, which I agree is totally different from a cartoon, it's really trying to make you often think that it's the person. And a lot of voters probably are likely to believe that in a way that they are clearly not if there is a sketched out cartoon. And then of course, course any AI generated misinformation. AI generated avatars on social media, the kind of stuff we've seen in past elections, but more sophisticated, cheaper, easier to produce at scale. So that would be one. And it does seem that at least here in the United States, the only regulation that is forthcoming would have to do with disclosures, disclaimers. It doesn't seem like we're likely anytime soon and maybe ever because of First Amendment concerns to regulate the actual creation of the material. So that's thing number one, obviously concerning then number two would be AI use not for deception of any sort, but for persuasion, which is what Kevin was talking about at length with that study. And I guess there it feels to me like it's also sort of an escalation of stuff that we've already seen in recent elections. There was a lot of hullabaloo around micro targeting of advertisements, which was way more sophisticated than what people had been able to do before when it came to getting very specific messages in front of specific people. And now it seems like, well, you're going to be able to probably do that a lot more easily and you're also going to be able to do it more persuasively because the AI is going to be tailoring the messages themselves. And then I suppose for me the question question is how dangerous is that and why is it dangerous and why is it different than what we've seen before? Why is it not just better persuasion? I mean, we've always been trying to be as persuasive as possible. Political parties have been. One reason would be, which again Kevin touched on that if AI is biased, that bias is going to make its way in. Another reason that Kevin also mentioned would be if your ability to persuade depends on having access to the best AI, then maybe you need a bunch of money to do it. And that disadvantages the little, little guy. Although to some extent having access to AI at all makes it easier for small campaigns to develop the ability to persuade larger audiences. So that would be one, two that I've been thinking about and that I was reading and I wish I had the name of it, but I was reading an interesting paper about when I was looking into all of this would be kind of the idea of a sort of feedback loop. So I know that AI could be really useful when it comes to becoming a more educated vocabulary. You could ask what does this candidate think about the issue? And get a much better answer. But I know already people are doing that and that in a combination of kind of who should I vote for based on my beliefs? And AI will give you an answer. Sometimes you might think it's good, sometimes you might think it's bad. Probably depends on how high information you are or aren't in the first place and how well you know yourself. But if parties are going to start using AI to inform their strategies, either their strategies generally or their strategies for specific populations they want to win over over, and then voters are also going to be talking to AI to help figure out who they're going to vote for, then that's going to inform what AI tells the parties to think about. And then it's all just going to spit back and just be this loop that's like an AI generated loop. And I think that's concerning because again, it raises the question of where's the information coming from? Is it only going to be, is it going to be forward looking at all ever? Is it going to be organically generated from the new human thoughts ever? And then the final thing I'd say in this long rambling discourse is that I think generally if you're relying on AI a lot, either as a party or as an individual, to inform your decision in an election that is to a certain extent going to atrophy your critical thinking skills. And a populace having good critical thinking skills is important in a democracy.
- [51:21]
Scott R. Anderson
Yeah, I think this study is really interesting, but the fact that it boils down to AI's comparative advantage primarily over an educated trained individual is rapidity and volume is interesting because to an extent you're like, yes, but it kind of makes sense maybe when trained against our expectations currently. But will those expectations remain static when multiple people start? When we start engaging it all the time and the throughput of data and facts is so constant and it's particularly telling. I think that again, this is why I want to spend more time with this study. If it's debating policy issues and political issues where they're kind of a substantive debate and their previous political valences, but somewhat more technical and technocratic, I wonder how it translates into voting persuasion like actual candidates, where we know a lot of what people vote for is less about strict rationality. The reason we hear authenticity raised so much with candidates these days is because so many voter key swing voters make a decision based off of very subjective senses about a candidate. And no doubt AI can play a prime role in framing those. But particularly if you restrict AI to being honest. If you limit the duplicitous uses of AI, which is hard, but there are ways to do it. I'm curious to see how that comes into play. I'm not sure it's actually not healthier to be able to give people a lot more information. But Kevin, I want to bring in one last element before we depart from this because we talked about about. I think Molly did a great job drawing out those two buckets of use of AI. The deceptive, which you know, is just misleading things and the facts, which is really presenting a broader array of facts and ideas. There is a third use that Molly hinted at, but I think is worth breaking out. That's the. For lack of a better way to put it, I'll put the oracular or maybe the consultant element of it, which is you ask AI, you put inputs into the black box. Hey, you know, who should I vote for based off xyz, based off what I think? Or you have a conversation with it and it spits out an answer.
- [53:17]
Roger Parloff
Answer.
- [53:17]
Scott R. Anderson
That's one case where you really, because of the black box element of the AI might be particularly concerning because you're not giving a position to advocate for you are. And then having it bring up the best arguments. You're really asking some fairly subjective senses. How does that fit into this sort of equation?
- [53:33]
Kevin Frazier
Yeah, well, first I just want to echo Molly's point on the fact that we can design AI tools that can actually improve and hopefully enhance our capacity to be better democratic citizens and to. To have a more deliberative discourse. As it stands right now, based off of studies like this, I presently am more worried about what I'd. I'd refer to as Democratic De Skilling. And this relates to the fact, as you're pointing out, Scott, which is we saw a recent study from the Washington Post where they tested various models against different political questions to get a sense of the ideological leanings of models. And the majority of the models tended to skew toward the left end of that spectrum when answering certain policy questions. Gemini, per the Washington Post, was neutral in a lot of contexts, but as
- [54:23]
Scott R. Anderson
the Washington Post and slightly sycophantic in
- [54:26]
Kevin Frazier
my experience but critically, as the Washington Post article notes, as flagged by an excellent scholar, what does it even mean to be neutral? Neutral in and of itself is a stance in our political context. And so, so knowing that that information isn't presented in the same way as, let's say, a crash test safety rating on a car, you know, what kind of car you're getting into and driving and leaving the lot with, when you go ask these sensitive political questions to Gemini or Claude or ChatGPT, you may not have a sense that you are slowly being directed or steered towards certain perspectives and views. And that, to me, is the, the sign of that sort of Democratic deskilling that I'm, I'm worried about, about where people begin to start to just acquiesce to whatever the model saying. We know from Pew studies and from the Gallup polls, for example, that while a lot of Americans don't trust AI companies, they do trust the AI. And so that sort of willingness to trust the response, I think is particularly worth studying. Now, just to nerd out for, for one second, because I am trying to make fetch happen here with this idea of the future of governance studies, which I think is something that we should all, as scholars, be interested in, which is to say, what does this mean?
- [55:49]
Scott R. Anderson
My program at the Brookings Institution, in case anyone's doubting, I don't think that's exactly what you're talking about, but. But we're a vested interest as well.
- [55:55]
Roger Parloff
There we go.
- [55:55]
Kevin Frazier
There we go. What does this mean for things like ballot initiatives and referendum? Should we have something like AI pair parody, where if one party or one sponsor gets to use the latest and greatest models, do we need to make sure that other parties have access to that, those same AI tools? Because ultimately, when we think about how this may play out. And again, to riff off of Molly's excellent rant, now I'm going to go even nerdier. I think the guarantee clause needs to start being invoked here because we are talking about Republican governance, and Republican governance is contingent upon deliberation. If we don't have meaningful deliberation and discourse in our democracy, then I'd argue we're moving away from a Republican form of government, which could be a basis for having more conversations about the guarantee clause. So whoever wants to nerd out about the giant provision laying dormant in our Constitution, give me a ring and I'd be happy to nerd out with you.
- [56:56]
Scott R. Anderson
Interesting. It makes me slightly nervous given some of the uses we've seen made of the guarantee clause in the last year or two. But, but fair point. There's really a broader concern about Republican government and Democratic government that comes into play here. But we are not done with topics today. We have one more we need to spend some time on, so let us shift our focus to that. We have another piece of litigation in the pipeline now, as we discussed last week, we know Anthropic's Fable model, the kind of sanitized, guardrailed version of the Mythos model made briefly publicly available to those Claude Pro subscribers among us for a brief, beautiful 72 hours or so. I mean, not even that long, I don't think 48 hours that the kibosh indirectly, but probably not unexpectedly put on it by the Trump administration two weeks ago or a little less than two weeks ago, where they essentially delivered an export control is informed letter saying, hey, you can't share this with foreign nationals because that kept Anthropic from even sharing it with its own foreign national employees who are essential to maintaining it and maintaining its safety. According to them. They said, okay, we just have to pull the model altogether. And they've pulled and it's remained pulled. And if you go on Claude now, you'll see not only is it pulled, but you have a little info window saying it's gone. Here's question. Why? And it goes to a little explainer that maybe throws a few fairly polite but nonetheless pointed punches at the administration for what it's doing. Now we're seeing the first lawsuit over this. There's been this big question as to the administration's ability to impose this sort of expert control, its legal authority to do so. And now we're seeing a first lawsuit over that from a tech startup that was leaning on Fable, or at least would like to continue to lean on Fable to develop its product and deliver its services. Kevin, let me turn to you on this. Talk to us a little bit about this lawsuit. What jumps out to you about it?
- [58:34]
Kevin Frazier
Yeah, so this is a dispute between Legion Legal Tech, which I hadn't heard of previously, but always glad to know of a new AI legal tool, and the administration. And the challenge here is really based off of the fact that Legion Legal Tech, to your point, is of kind. Scott claims to have been relying on Fable 5. I'm not sure how much reliance you could have established in that beautiful 72 hour period. But they are emphasizing that Fable 5 was indeed in their workflow and being used in a business context. And so they are seeking a end to this action and the fact that we have this prohibition on the ability of foreign nationals to access Fable 5 and Mythos 5. And by extension, the world has lost access to Mythos 5 and Fable 5. So they are challenging the administration's actions on three grounds. First, the idea that there has been an excess of the Export Control Authority allocated to the administration under the ECRA and by extension under the ear. So I'm sure we'll get into that in a second. The second would be a challenge under iepa, everyone's favorite emergency law provision, and the Berman Amendment that is related to ipa, about when IIPA can be used and when it cannot be used with respect to information or information materials. And then third, we have a potential APA challenge arguing that the action was arbitrary and capricious. And so to just walk through those three in slightly more detail, fail. I'll flag that there's a lot of nuance, and I love that we'll be able to get Scott's hot takes on the ecra. But when it comes to the idea that this is a action that the administration was authorized to take, or to be more precise, the BIS was able to take by issuing the as Informed letter here, we're, we've seen a long debate as to whether an export covers something like accessing a model. And in this case, we're seeing Legion Legal Tech challenging the idea that you can apply a control to something like this, accessing just a model, given that traditionally an export has referred to something that is tangible or something that has been explicitly added to the control list. And in this context, we haven't seen that be the case. In fact, Congress is currently debating legislation that would kind of explicitly allow for export controls related to things like remote access to software. And so the challenge here would be, look, the application of the ECRA is not suited to this particular context. Therefore, this is an ultra virus action, which basically just means that there's no legal authority for the administration to be acting in this regard. Likewise, under the second count, we see that if this is an action that's being justified under ipa, an exercise of emergency power statutes, IPA is contingent on recognition of a sort of particular national emergency and a particular threat to national security that emerges in part or wholly from outside of the country. And here there has been no such declaration that this threat is emerging from beyond the country or predominantly from outside the country. Anthropic is a US Company. The model is generally housed within the us so how IIPA would apply here is another challenge that's being based off of a, or allegedly an ultra virus action. And then finally, if There is the ability to challenge this decision under the Administrative Procedure act, the apa. The argument is that this is arbitrary and capricious. That one, I think, is going to be a little bit harder to reach because there are a number of bars on what can be challenged under the APA with respect to export control decisions, but that's just a quick lay of the land. I think that it's important to note how hard it is to win in ultra virus action. You are going to need to really have a slam dunk case when it comes to making sure that you are showing that the administration has acted without legal authority. So the standard here is that the action must be entirely in excess of delegated powers and contrary to a specific prohibition that is clear and mandatory. So entirely in excess of delegated powers, just hanging on that language in and of itself, itself, that is a difficult standard to reach. And so that's going to be the most difficult part, in my opinion, for Legion Legal Tech.
- [63:26]
Scott R. Anderson
Roger, what do you make of this complaint? I think Kevin laid out the key arguments very well, including some of the potential issues with some of them. I'm kind of curious from your, you know, the perspective somebody looks at a lot of litigation around this, like, what's motivating this? And like, plaintiffs come with these lawsuits a lot of motivations. Obviously, they like to get mythical and fable back. I'm kind of curious about how strong you think these arguments are. And insofar as they may not be the strongest, I may be leaning a little bit towards that direction, but I can explain why in a little bit. Why do you think the motivation may be for pursuing this sort of lawsuit?
- [63:59]
Roger Parloff
Well, I mean, what they say is that the harm is existential for them, irreparable and immediate. As Kevin pointed out, they say that Fable 5 is crucial to their development process. And as Kevin pointed out, it was only in existence for three days before it was taken off the market or forced off the market. Now, that could be just the nature of the game. Now that a crucial AI tool is that important, that, you know, he talks about the plaintiff, talks about the blistering pace of competition right now, and. And a tool like this is a game changer, and you're competing. And it's essential. The judge here is going to be Richard leon, who is 76 and George W. Bush appointees. Good judge, but he's not Kevin's age, and he may. Look at this.
- [65:03]
Scott R. Anderson
You don't know how old Kevin is, Roger.
- [65:06]
Kevin Frazier
My nickname in college was Grandpa. So you know who.
- [65:09]
Roger Parloff
Okay, okay. Okay. But, you know, it may be hard for him to think that, you know, a product that's been in existence for three days could become that crucial that you need a preliminary injunction. And interestingly, they did not seek a tro, a temporary restraining order. And then, as Kevin also said, the standards for proof for a non statutory ultra virus action are extremely high. I think the Supreme Court has referred to it as a Hail Mary action. So all of those. It's not an easy case. Another thing on the merits is that the accusation is that there's a jailbreak vulnerability here. And the response from Anthropic has been, well, it's a narrow jailbreak, and that's sort of fuzzy for especially a judge in my age bracket to understand. But there's fascinating things about this suit, like the declaration from the CEO, the way he speaks. He's an engineer himself, Arthur Rothrock. The way he speaks about Fable 5, it's just like. I mean, it's so. The terms are so human. He says things. Let me just quote. In my direct experience, Fable 5 was markedly better than the other models we use at understanding our code base as a whole and at planning and executing large, complex feature build outs. Reliable, it said its willingness to sanity check ideas and push back on questionable engineering decisions. He cites. He says we need a model that functions as a genuine critical thinking partner rather than one that merely validates whatever it's told. I mean, it's so amazing to see the personal terms that he describes this partner that's been taken away from him. I mean, I don't know how Judge Leon will respond to that, but it's an uphill battle, but it's sort of a fascinating one.
- [67:34]
Scott R. Anderson
Yeah, I agree, but I do think. I mean, this is where I come out on this. This is a highly symbolic filing. Like, this is a message sending filing. First, like bringing AIPA into this does not make any sense. It gives you a really convenient reason to talk about the Berman Amendment and talk about the First Amendment Amendment, which, frankly, there is this, like, borderline obsession in the technological community that software is speech. And therefore, which is a very broad reading of a very narrow decision by a very lower court that does not actually carry the weight. I think a lot of people think it does, but has a strong rhetorical weight in a particular community. My sense that this is leaning into this, and this is a lawyer's effort to make that sound as reasonable as you can by pulling into the iipa, not totally unrelated. IIPA was used to enforce export controls for years and years. And years when there is a lapse in export control statutory authorization. That is no longer the case. That is the ECRA enacted five or six years ago. I can't remember exactly when off my head. 2018 was it? Yeah, we wrote about an offer at the time. I remember that when it was enacted, that provided new statutory authorization. There isn't a Berman amendment applicable. There are other restrictions, some of which have similar kind of speech concerns, but nothing that general or specific. So when you bring that in, you really are finding a reason to talk about it. And they don't even really say why they think AIPA is involved. They're 100% right. You would need a national emergency declaration. There hasn't been one. Lots of executive orders about AI. I'm not aware of any national emergency declaration which you would need for ieepa. So the predicates just clearly aren't met. Why would you waste your time, your complaint, talking about it? Because you want to talk about speech and throw a bunch of stuff against the wall. The ultravirus action is kind of a concern. And for any of the complaints about the export control, I mean, there is a colorable argument you can make about how far the export control regulations can go. But the use of IS informed letters isn't new even in the AI space. They've been used for chips and semiconductors. It is a little news applied to this. I think there's a lot of established practice around the idea that cloud services are not something that's generally subject to this. I'm not sure that is a hard required reading of the statute. I don't think it is really, at least by my preliminary reading. But I haven't spent enough time on this to come up with a firm opinion on it.
- [69:40]
Kevin Frazier
On it.
- [69:40]
Scott R. Anderson
But I think the key point is here the argument is if your grounds is you need to have really unimpeachable sense that you are acting outside the scope of the statute. I think that's a really high bar to meet in this context when other informations and services are regulated under export controls. And there are lots of parallels you can draw to this, to things that are regulated, even if you might be able to distinguish in certain grounds. And there are some policy determinations that cast some doubt as to whether traditionally you would view these things as regulatable under that sort of lens.
- [70:11]
Kevin Frazier
Scott, to what extent do you agree with the characterization in the complaint that we saw here, the administration effectively skipping steps that usually would be taken before a as informed letter is even issued? To what extent do you Think that may be persuasive. My, my quick read or my sense of it? As we've talked about and as Roger written raised, we're seeing that if the standard is this Hail Mary standard of ultra virus action as you just outlined, Scott, if there's an argument to be made about how to interpret the statute, you're going to lose in this context.
- [70:47]
Scott R. Anderson
Yeah, I think that's probably right. And I'm just not sure that their sense of the prerequisites for the as informed letters is quite accurate. I mean, the as informed letters are an extraordinary measure, basically saying, hey, if something is within the scope of our regulatory authority, it gives a fair amount of, of leeway to the executive branch to say, hey, we're going to send a letter saying, well, this specific product is subject to this. You've got to stop doing this. The whole idea is for it to be a kind of emergency measure to say, hey, we need to move quickly on this. And so that belies some of this assumption that you're going to have a lot of these strong process behind it. Now, again, I want to look at this and I want to think more about are we confident that this is within the scope of what can be regulated in the scope initially, but I don't know. I'm not convinced by the rest of this complaint that that argument is well founded because the rest of the argument arguments I don't think really are. Now why are you doing this? I think is really a sign that the industry really wants this. Like a lot of people really did like this product. I have to say I use Fable for a hot second and I was quite impressed. I do a lot of like vibe coding projects for a bunch of data and research and scraping and I was like a substantial. Felt different when you used it in a way that was impressive.
- [71:55]
Kevin Frazier
72 hours of bliss.
- [71:57]
Scott R. Anderson
Yeah, exactly. I mean, I was still very impressed by Opus 4.8, which I think is the one I usually end up using. It's great. I'm not going to say I'm like torn up on the inside about not having access to Fable, but I can see if I were a business that really relied on this for much more sophisticated stuff than I'm doing, why I would want it. And I think it's just a sign there's going to be a lot of pressure to get this out there as more industries begin to feel the real competition are going to say, particularly when you see frankly, OpenAI maybe stepping up with their own product and potentially foreign competitors entering into the space. There's going to be this constant drive to say, well, we need access to the latest model that we can get. And once you give people a little taste of it, it's going to be hard to kind of pull that back in because they're going to sense about what it could do for them. I don't think there's lawsuits going anywhere on that ground, but it's a useful messaging device. It's a useful message to say, hey, we want access to this and we're going to keep pushing to get it. And I suspect we're going to see more chatter in this domain, if not other lawsuits asserting similar things, even if the legal avenues don't seem super plausible. Because a lot of this is about messaging and, and putting political pressure on the officials and to some extent on Anthropic to find ways to work with Executive Branch to get these products back to their subscribers.
- [73:05]
Roger Parloff
Roger, there's something I'm not sure about as far as I had a different reading of when I saw the arguments about IEEPA in there. I don't know how much is known for sure about the bases for the government's basis for this, for shutting Fable down or for ordering it off the market. And I thought that possibly the plaintiff here was just saying, if it's based on the Export Control Reform act, then here are arguments. If it's based on ipa, then here are our arguments. I thought it was partly just on his part, not knowing what the government was going to come back at them with. But I would say that whatever the. Obviously one of the strengths of his position is that the suspicious backdrop of all of this, the government's hostility to Anthropic over several months. But another complicating factor is that here they seem to want to. And maybe this is just a way to try to carve out a small, equitable cocoon for themselves that won't create a big problem with the law. They want to say that their developers happen to be Canadian. And so they want to say, well, so these are foreign nationals of an ally as opposed to foreign nationals of an adversary nation. Making those distinctions is not all that easy, especially today. It's also only like it comes down to it. It seems to be three people as a small company. There's two developers in Canada, they're expecting a third later this month. And then there's him, the CEO. So it would be a relatively small. You could craft a preliminary injunction that was pretty narrow, but I don't know. It's a funny case.
- [75:08]
Kevin Frazier
Well, to piggyback on the earlier point, though, about the jailbreaking here, I do think it's really important to note that the, the, this kind of jailbreak that was allegedly discovered by Amazon is something that's just inherent to the models at this point. You don't see that it's technically feasible to eliminate each and every jailbreak. What's critical is that this wasn't the sort of universal jailbreak that would all of the sudden allow Fable 5 to be used by any bad actor for any purpose. And what's really interesting, nerding out for a second on folks like Lon Fuller and Waldron and other awesome Rule of Law scholars dollars, right? The basics of the rule of law. When we say that phrase, we're referring to generality and prospectivity. Minimally have to be a part of the rule of law, which is to say the law should apply to everyone in the same fashion and people should know what the law is in advance of their conduct. And here, just to zoom out at a bit of a higher level, neither of those has been satisfied because every model has the sort of jailbreak that is allegedly the basis of all of these concerns around Fable 5. And secondarily, whether or not, you know, the as informed letter was applied as intended or as statutorily authorized, this shocked everyone. No one saw this coming, let alone at 5:12pm on a Friday during a World cup game. That is not how we think through the rule of law or what we would like to see in this process. And so we do need Congress, just for any staffers listening, Congress, please act. And please develop a system that's clear that everyone knows how a model is going to be evaluated so that we don't have Canadian legal tech companies who are angry about how and when we
- [76:49]
Roger Parloff
enforce our laws just to respond on the jailbreak. And that is, of course, a legal tech's argument, if that's its name.
- [76:58]
Kevin Frazier
Legal Legion Legal Tech.
- [77:01]
Roger Parloff
Legion Legal Tech. But what I think Judge Leon might wonder about was if it's so minor, why did Amazon go to the White House over it? And Amazon is an investor in Anthropic, and it's also a commercial platform. It's enormously alight. That will puzzle him, as it puzzles me.
- [77:26]
Kevin Frazier
Puzzles me too. Andy Jassy, come on to scaling laws. At any point you have an open end invite.
- [77:32]
Scott R. Anderson
There we go. Well, there's a lot more to this story that's still going to unravel, and I think this litigation is going to be a part of it. If perhaps a Smaller part of it, but we'll see because we have more to come on these fronts in the weeks to come. But we are out of time for today's episode. But of course this would not be rational security if we not leave you with some object lessons to ponder over in the week to come. Unfortunately, Molly had to step away before the end of this week's episode, but she sent in her object lessons lesson separately. So let us turn to that now.
- [78:00]
Molly Roberts
Okay, so for my object lesson, I am going to take advantage of the fact that I am recording this later than you guys to not adhere as closely to my self imposed requirement that I bring a physical object as usual and bring something that I can't physically bring because it's too big. But it is in my home. Home. It's a series of posters created by this Swiss graphic designer, Eric Nietzsche, and they are General Dynamics posters created in the post war period. They're part of a campaign called Atoms for Peace where General Dynamics was trying to convince the world that actually atomic energy isn't just about blowing things up. They're about good uses and uses that are in the public interest, which I feel is certainly applicable to the AI conversation that we had this episode. And also they're just visually pretty cool. So I recommend that people check those out.
- [79:05]
Scott R. Anderson
Kevin, what did you bring for us this week?
- [79:07]
Kevin Frazier
Yeah, so my object lesson is a nice little beach read. So I highly recommend that folks check out Gillian Hadfield's book Rules for a Flat World World. It's an excellent read for anyone who wants to nerd out about what is the purpose of the law, where do we derive law from, and what sort of norms and evolutionary processes go into how we govern society. I think it's a really fun read as a lot of folks are analyzing as we nerded out about Scott, the future of governance and how we're going to see legal systems evolve in this fascinating era. So, so definitely check out that book. Professor Hadfield is a really smart and innovative thinker on these issues and I just think we need lawyers who are asking bigger, bolder questions and she's certainly doing that.
- [79:59]
Scott R. Anderson
Wonderful, wonderful. Well, for my object lesson I'm going to do a little bit of a cop out. I think Ben took one of these a couple of weeks ago. I did not come up with a good object lesson for this week because I'm about to go on vacation. So my object lesson this week is that there will not be irrational security next week because I'm going to go on vacation if I'm be honest, my brain went ahead and went there a few days in advance as it's prevented me from recalling the object lesson I thought of an hour and a half ago at this particular moment that I failed to write down. And for the moment I'm going to pass on that. But I hope everyone enjoys their Fourth of July week, enjoys the 250th anniversary, even if you're not so bright here in Washington D.C. which I am getting out of town for a reason. Perhaps other residents are as well. Well, but we will be back with Rational Security the week after the 4th of July, our 250th plus one week anniversary episode. So keep tuned for us then, but otherwise enjoy your holiday. Roger, bring us home. What do you have for us this week?
- [80:49]
Roger Parloff
A friend, actually a family member, gave me a book called warriors of Winter by Olivier Norek. It's about the war between the Soviet Union and Finland in 1939. It's about lasted a little over 100 days, something I knew nothing about and a very brutal war with a hero, a sniper for Finland who apparently single handedly killed more than 500 Soviets or people fighting on behalf of the Soviets. But it's very reminiscent of the Ukraine war. You know, it was David versus Goliath situation and the fact that they held out from more than 100 days shocked the world at the time and probably was one of the reasons that Hitler abandoned the treaty with the Soviet Union. I think he thought they were weaker than he realized. But it's quite a fascinating read about something I know nothing about. A really heroic stand.
- [82:01]
Scott R. Anderson
Yeah, really fascinating. Really fascinating. Well, well that has a couple of great beach reads that I will be veiling myself myself of perhaps next week when I am on a beach in a lake. Though not real beach, but whatever, I'll take it. But until then we are out of time. That brings us to the end of this week's episode. But remember, Rational Security is of course a production of Lawfare, so be sure to Visit us@lawfaremedia.org for our show page, for links to past episodes, for our written work and the written work for the Lawfare contributors, and for information on Lawfare's other phenomena phenomenal podcast series. While you're at it, be sure to follow Lawfare on social media wherever you socialize your media. Be sure to leave a rating or review wherever you might be listening and sign up to become a material supporter of Lawfare on Patreon for an ad free version of this podcast among other special benefits. For more information visit lawfaremedia.org support our audio engineer and producer this week was me of me and our music as always was performed by Sophia Yan. We were once again again edited by the wonderful Jen Pacha. On behalf of my guests Molly, Roger and Kevin, I am Scott R. Anderson and we will talk to you next week. Until then, goodbye.
- [83:11]
Chase Sapphire Advertiser
This year's girls trip to Telluride was the best. We one upped ourselves with my Sapphire Preferred card and with 5 times points on Chase Travel plus 3 times points on vacation homes with top brands, we got this incredible cabin. It was a mansion and with three times the points on dining, we ordered a Wagyu steak dinner and that pistachio gelato was too good. So where should we go next year?
- [83:32]
Scott R. Anderson
I've got ideas.
- [83:34]
Chase Sapphire Advertiser
Chase Sapphire preferred the card that's preferred for a reason. Cards issued by JPMorgan Chase bank and a member FDIC subject to credit approval terms apply.