Who’s liable when AI goes rogue?

Dave Bittner: Hello everyone, and welcome to "Caveat", N2K CyberWire's privacy, surveillance, law, and policy podcast. I'm Dave Bittner and joining me is my co-host Ben Yelin from the University of Maryland Center for Cyber, Health, and Hazard Strategies. Hey there, Ben.

Ben Yelin: Hello, Dave.

Dave Bittner: On today's show, Ben discusses a new bill in Congress that would impose liability on AI developers. I've got the story of AI-generated testimony that triggered resentencing. And later in the show, my conversation with Garfield Jones, Senior Vice President of Research and Strategy at QuSecure and former CISA Homeland Security Executive for Quantum. He's sharing some insights on migrating to post-quantum cryptography. While this show covers legal topics and Ben is a lawyer, the views expressed do not constitute legal advice. For official legal advice on any of the topics we cover, please contact your attorney. [ Music ] All right, Ben. Why don't we kick things off here? What do you got for us this week?

Ben Yelin: So, ever since the Hugging Face incident this summer, I think there's been a conversation about how do we hold AI developers liable --

Dave Bittner: Yes.

Ben Yelin: -- in civil law? Because if they're going to be presenting what they admit is an existential risk to humanity, they should at least bear some of the costs of those potential harms.

Dave Bittner: If my dog bites you, it's my responsibility. Ben Yelin: Exactly. And I can tell you, we've had this dog comparison to AI for years.

Dave Bittner: Is that right?

Ben Yelin: It's very similar.

Dave Bittner: Yeah. Ben Yelin: You have somebody who actively trains this non-sentient human and that thing goes out and does things at your behest, there's some similarities there. But we finally have a proposal in Congress. This is entitled the, and this is, we've got a great acronym here, Clear Liability for Artificial Intelligence Misconduct, or CLAIM, Act.

Dave Bittner: Oh, nice.

Ben Yelin: And this bill would hold AI developers legally responsible when their autonomous agents cause harm through conduct that would, and this is the key here, be unlawful if performed by a human. So the simplest example is hacking, right? That's criminalized under the Computer Fraud and Abuse Act. So if an AI agent does it, liability imputes to the developer as if a human had done it.

Dave Bittner: So, the Hugging Face incident would have had legal peril.

Ben Yelin: Yes. So, OpenAI would have been held liable, even though you couldn't go in and prove that OpenAI was not only negligent, but that there was a defect in the product. There are a couple of kind of conflicting standards that we use for liability when it comes to torts and civil cases. Most of the time we use negligence. You have to prove that somebody breached their duty of care, that there was causation, and that there's some type of injury or damages. There's also something called strict liability for products. And I think a lot of people have brought up this idea around AI, that there are some products that, you know, we don't have any visibility into the sausage making, that it would be too difficult for us to go in and prove that a company was negligent. Like, if you showed me every single thing that OpenAI did when developing the model that hacked Hugging Face, I, A, wouldn't be able to read it without AI summarizing it for me.

Dave Bittner: [Laughter] Right.

Ben Yelin: And B, I just would simply not understand it.

Dave Bittner: Yeah.

Ben Yelin: So, strict products liability, you can impose liability even if you cannot allege negligence. Obviously, the big developers are going to be against this because they would face significant liability if there is any harm. So, this kind of splits the apple a little bit where it's not necessarily strict liability in the sense that everything that goes wrong with a model you can impute to the developers. Instead, it's just making the developer the human for the purpose of the intent requirement in both civil torts and in criminal law. So, AI developers would be liable when their systems injure third parties through actions that would constitute negligence, intentional tort, or a crime if performed by a human. There is a presumption of intent. So that's the key here, that obviously we cannot impute intent onto these AI agents. That's part of the problem. This law would presume that the AI agents have the requisite intent to fulfill the necessary elements for the tort or the crime. It would impose liabilities regardless of reasonable precautions. So, I think a lot of the developers would say, you know, maybe we can have some type of liability shield if we do red teaming, if we take security measures.

Dave Bittner: Right.

Ben Yelin: And this proposal would not allow them to do that.

Dave Bittner: Would something like this be EULA proof?

Ben Yelin: Absolutely.

Dave Bittner: Okay.

Ben Yelin: Yeah, I mean, I think a EULA would not get anybody out of a law like this just because of how broad it is and the overall intention. There are a couple of exceptions in here. So if a user of the application does something that causes the harm, that would be kind of the intervening act that would shield liability on the part of the developer. So like if I, for example, used one of these models and engineered it in a way so that it committed a cyber attack, I would be held liable and not the AI developer because I committed this type of intervening act. Another thing that the developers are going to hate, this would allow for a private way of action. So anybody injured by AI agents could sue for damages and other relief. I believe there would be a lot of lawsuits, because I don't know if you've heard, but these, there's a lot of money in these companies, especially even though they're not really making a profit right now, you know, they're going to go public presumably pretty soon. And at some point there's going to be a lot of money and people are going to have incentive to sue them. The last thing is that this bill would not preempt state liability laws. This would be a floor and not a ceiling. States would be free to regulate and impose liability above and beyond what this legislation does. So this is proposed by Representative Lori Trahan, a Democrat from Massachusetts. It is more of a proposal designed to start a discussion about this. We're running out of time in this Congress. There's going to be, you know, just a few weeks of a lame duck session left this year. I think this is something, if it were to move, would be more likely to move next year. But I think it's an indication of what these companies are going to face from policymakers, just given the high-profile nature of these incidents.

Dave Bittner: Why do we need this? Why is current law, the Computer Fraud and Abuse Act, for example, inadequate?

Ben Yelin: Well, for one, that only applies to crimes. And so there are going to be some incidents where there are civil damages and tort, whether it's negligence or an intentional tort. We can't use negligence because that's too hard to prove in a scenario where these agents are acting inside complete black boxes. We have no understanding of what they're doing, what their motivation is. So there would be no possible discovery process to see if they were negligent. And generally, even in products liability, you have to prove that there was some type of defect in the manufacturing of that product. And same thing here. Like, I would have no clue how to prove it one way or another. And the average person shouldn't, you know, have that obligation to prove whether something happened in the creation of this product and the creation of whatever AI system that's being used that caused the resulting harm. Like, that just shouldn't be a user's responsibility or a third party's responsibility. And if we don't change our liability laws, all of the harm will fall onto the victims. Anybody who suffers as a result of the act of an AI agent will pay, well, they're the entire cost. And as a society, I think that's an outcome that we don't want. That's why we have these laws in the first place, so that the harm is distributed among the people who caused it. Even if the victim was partially negligent, you know, in most cases the victim can still collect because the defendant is going to be more at fault. And so we want a system in which the burden of the harm doesn't fall entirely on a single person.

Dave Bittner: It's interesting to me that House Speaker Mike Johnson seems to be behind this, or at least has expressed support, approval, which to me would not align with where I would expect that the president would go.

Ben Yelin: Yeah, I think he was more just saying, like, it's good that Congress is talking about this issue.

Dave Bittner: Okay.

Ben Yelin: I think when push comes to shove and this went on the House floor, you know, this would be a pretty burdensome regulation. Obviously the companies would be against this. And you're absolutely right, this goes against the philosophy of the administration, who, instead of imposing these types of regulations, wants to have kind of a, as President Trump called it, a moral pact between the big AI developers. So, they had this big summit at the White House. All of the major players were there. They agreed to pace the frontier and take all of the safety measures. And, you know, from the President's perspective, that is, that should be in lieu of Congress coming in and imposing liability. And there would be significant consequences to this. I mean, I think the big developers would be much less likely to take risks, much less likely to develop new models with advanced capabilities because they'd be very concerned about facing this type of liability. So, you know, they would go to Congress and say if you pass this, you're going to kill our industry. And maybe that would be an overstatement, but I, you know, I think there's some truth to it. Like, it really would inhibit the ability of these companies to take major risks and get away with it if there are unexpected harms.

Dave Bittner: Can you think of any examples through history of where, you know, some developing technology had to be held back like this? I'm trying to think of something myself.

Ben Yelin: Absolutely. I mean, our tort law has developed because we keep having these new products come online that present different types of damages. So, a lot of our negligence jurisprudence at common law came about through railroads. There were a lot of accidents that happened just in the course of building railroads, people as passengers. Then we had the industrial revolution, and that really was the impetus for products liability, because we started building all these products in factories. People don't have visibility, just normal people who might be injured by these products don't have visibility into the process that goes on at the factory. So, they would be losing all these cases because they'd be like, these guys were negligent, and the guys would say, prove it, and the person wouldn't be able to prove it. So, the upshot of that is that anybody who was affected by one of these products, pretty much no matter what, was going to bear the entire costs of whatever injury befell them. So if a Coke bottle exploded in their face, you know, that ends up being their responsibility. The manufacturer bears no responsibility. And from a policy perspective and from a legal perspective, that seemed like a bad outcome, and that's where we developed products liability. We've seen this for, there's a doctrine in law called strict liability for abnormally dangerous activities originated in a British case that everybody learns about in law school, Rylands v. Fletcher, where somebody had a reservoir on their property and water dripped and caused the release of coal or something.

Dave Bittner: Oh.

Ben Yelin: And the doctrine there is that if you're doing something that's so particularly dangerous that it's abnormal for the area in which that activity is taking place, then you are strictly liable for any harms that come from that type of activity. So, that's another theory that I think could potentially be very relevant here, that if we see these continued attacks, cyberattacks from these agents, maybe this would qualify as that type of abnormally dangerous activity and we could just impute strict liability onto it. Some court might adopt that as a theory. Until they do, this is a very unsettled area of the law, so this proposal I think is just kind of seeking to clarify that.

Dave Bittner: It's kind of interesting to me that this is being taken up here, given the political will of the White House, maybe, and the Republicans who have signed onto this, maybe a sign that the view of the people who seem to be frustrated with things like Flock cameras and certainly wary of all this AI stuff, maybe their representatives are hearing that outcry.

Ben Yelin: Yeah, so just to clarify, like, this isn't actually moving. This is just an idea [Laughter].

Dave Bittner: So don't get excited.

Ben Yelin: Don't get your hopes up, okay?

Dave Bittner: Right, right.

Ben Yelin: You know, this is like the, this is just the start of negotiations, is the way I would look at it.

Dave Bittner: Okay.

Ben Yelin: There are other legislative efforts. This article mentions bipartisan legislation from Senators Hawley, a Republican from Missouri, and Murphy, Democrat from Connecticut --

Dave Bittner: Yeah.

Ben Yelin: -- which would impose criminal and civil liabilities on AI agent operators for hacking violations of the Computer Fraud and Abuse Act. That's more of a narrow piece of legislation. It's great that it has bipartisan support, but it doesn't go as far as this legislation because that only applies to hacking-related violations under CFAA. This applies to all harm that results from actions of autonomous agents. So, I think this is just kind of the opening bid, kind of the maximalist position on regulating these AI agents. And it can be a jumping off point for negotiations. And I think if the political winds change with the upcoming election and the president has to negotiate with a Congress that's more hostile, they might come to the table with this as kind of their bargaining chip and say, like, you know, if you want to protect the industry and if the industry wants to play ball, that's great. But if you don't, this is the type of liability that we are prepared to impose on you.

Dave Bittner: I see. Interesting. All right, well, we will have a link to that story in the show notes. My story this week comes from the folks over at 404 Media, and it's kind of an interesting one. This is, well, so there was a crime committed, a gentleman by the name of Gabriel Horcasitas.

Ben Yelin: Nailed the pronunciation there.

Dave Bittner: Oh, thank you.

Ben Yelin: At least I'm guessing.

Dave Bittner: Shot and killed another man named Christopher Pelkey. This was during a road rage incident back in 2021. So a jury convicted Horcasitas of manslaughter, and during sentencing, I guess this is a standard thing, victims are allowed to make statements before the judge to make their case of their, well, how they're affected by the crime, right?

Ben Yelin: Right, and that's the normal course of business, is you'll have a victim impact statement, especially with something like manslaughter. I mean, these can be the most emotionally wrenching parts of any legal proceeding, to see victims talk about the impact that this murder or manslaughter has had on them. But what happened here is, takes it to a whole other level.

Dave Bittner: Yeah, so at this sentencing phase, Mr. Pelkey's sister, who's a woman named Stacey Wales, she presented to the judge a video that had an AI-generated likeness of her brother, the deceased. And this AI-generated likeness expressed forgiveness toward his killer, and suggested that they may have even been friends under different circumstances.

Ben Yelin: Which is nice, I guess. I mean, this whole thing is so weird, but.

Dave Bittner: Yeah.

Ben Yelin: I mean, it seems like the video might have been done in a positive spirit, a spirit of forgiveness?

Dave Bittner: So the sister says that the words were written by her, that they were not AI-generated words, but just the AI-created this avatar of her brother to speak her words. It was in her brother's voice. It worked. The Judge said, I loved that video, and then went on to impose the maximum sentence of 10 and a half years.

Ben Yelin: And specifically cited the video.

Dave Bittner: Right.

Ben Yelin: Which is bizarre, in my view, but yeah, nevertheless.

Dave Bittner: Yeah. So this went to appeal, and the accused argued that this lifelike recreation gave the statement an appearance of authenticity and authority that carried undue emotional weight. So, there's our legal term, Ben. Undo emotional weight. This was new to me. Can you unpack that for me at all?

Ben Yelin: Yeah, I mean, I think this is very specific to sentencing. I'm not an expert in sentencing or sentencing guidelines or sentencing proceedings, but I think the idea here is you want to have, allow for some type of due process for the person who's been convicted of a crime, and you want a fair sentence to be imposed. So, you have to have proceedings that comply with that process, which I think is just generally a set of rules, and undue emotional weight to me is a way of saying this goes beyond the normal emotional impact that victim impact statements have on courts and judges, that this is something else entirely. You know, there isn't, as far as I know, anything explicitly making this illegal. Like, there's no statute that prohibits you from showing an AI-generated video as part of a victim impact statement. You know, the judge here seemed to like it and he used it as justification for the maximum sentence. The appeals court that's saying that this was improper, saying that this caused undue emotional weight, isn't vacating the sentence, or isn't vacating the conviction. So this person isn't going to have a new trial. It's just going to be a new sentencing proceeding. And it's very likely that in this case, I don't know if it's going to go back to the same judge, but the judge could impose the maximum sentence just under normal sentencing proceedings where you have humans giving impact statements. And so the decision from the appeals court might actually not end up having a big effect here. I think this could be precedent setting. Obviously, this is just in one state. This is the appeals court in the state of Arizona, but I assume this is something that other states are going to encounter. And now we have at least this persuasive authority that these types of videos can be unduly prejudicial and should not be shown at these types of proceedings. So, that aspect of it is interesting to me. I've also seen, you know, I, most, most of the time this comes up in the context of celebrities where you have a celebrity that's died and somebody creates an AI-generated video of them saying something. People send it to that dead celebrity's friends and relatives. And, like, in my experience, people who were close to the deceased just don't like this. Like, they don't like words being put in the mouth of somebody who's not there to assert that statement themselves or defend the statement. And I'm sympathetic to that viewpoint as well. That, like, yes, the technology is great. It can be used for positive purposes. I, just as much as the next guy, enjoy AI-generated videos of, you know, football players cursing at each other in post-game press conferences (laughter). But this is, in a way, kind of emotionally manipulative. Like, I can understand that. So I understand both perspectives here.

Dave Bittner: Yeah. The article in 404 Media, I'll quote here, they say, "The appellate court agreed and cited State v. Rose, a 2007 murder case where a man was sentenced to death for the murder of a police officer. During the victim impact portion of the sentencing in that case, the judge saw photos of the police officer's children at his graveside. The court said those photos almost crossed the line into prejudicing the judge." So that's what's at play here is, I guess what I wonder about is isn't the whole point of the impact statement to --

Ben Yelin: Right, to be, like, maximally emotionally --

Dave Bittner: Right.

Ben Yelin: -- effective?

Dave Bittner: Right, exactly. Yeah.

Ben Yelin: I think this is more about just having some guardrails on it. Like, there is a line somewhere between a permissible victim impact statement and a statement that causes unfair prejudice. It seems like from what this appellate court is saying is portraying events and statements that never actually occurred crosses that line. I don't know what else crosses that line. They seem to have this precedent case that you just cited, but I think this is a case where they say that line exists, that there are some things, even in the context of what can be a very emotionally weighty proceeding, are just not allowed because the prejudice it would introduce to the proceeding outweighs the benefit of having victim impact testimony in the first place. So, the line again is drawing, their drawing here is around portraying events and statements that didn't actually occur. At least that's something that's at least seemingly justiciable. Like, you can know one way or another whether something is AI generated for the most part still these days. I think it'll be easier for judges in the state of Arizona now to have that bright line and know that these types of videos should not be allowed at sentencing proceedings because it'd be unduly prejudicial.

Dave Bittner: There's another comparison in this article that caught my eye. They were saying that it took about 15 years and five landmark cases in the US that went all the way to the Supreme Court before photography was an accepted standard in the courtroom. And so they're comparing this with AI. This is a brand-new medium, similar to how photography was when it was new. And so we're in this mode of figuring out what is and is not in bounds.

Ben Yelin: Not to mention just video evidence in the first place. One thing that's interesting to me is that in a sentencing hearing, this is up to a judge and not a jury. When we talk about an actual trial, I think the potential for undue prejudice is greater because you have 12 laypeople, men and women, who are on the jury who might be susceptible to that type of emotional, I don't want to call it manipulation but just the weight of the emotional appeal. But you'd think that, like, judges could be trusted to be impartial and fair, even if, you know, they've been emotionally impacted by a video of the victim saying something they didn't actually say. I think this is kind of underestimating the ability of judges who are well-trained, who go through, you know, hundreds of these sentencing proceedings, to be impartial. You know, I understand that there is still that risk for undue prejudice, but I also think it's not like a jury trial where you really do have to worry about people who have no experience in the legal system being emotionally manipulated. Here, we're talking about judges who have a lot of experience in the judicial system.

Dave Bittner: You mentioned the possibility that this could go back to the same judge. Suppose that judge says, "Fine, take the AI out. I'm still giving him the maximum sentence."

Ben Yelin: That could very well happen.

Dave Bittner: Is there any other avenue for appeal after that?

Ben Yelin: I mean, you could, the criminal defendant could appeal and say, I'm suspecting that this judge is biased because of the original emotionally manipulated video. That would be a whole other separate proceeding. The judge could just go and say, I was not.

Dave Bittner: Right [laughter].

Ben Yelin: None of that was in the record of what would presumably be the second sentencing hearing. And they would say, the judge would say, I'm imposing this maximum sentence for X, Y, and Z, other reasons that are unrelated to this video. I think that's probably what's going to happen if the same judge is presiding over that proceeding. It doesn't say in this article, it just says that there's, hasn't been a date set for resentencing. So traditionally, I, in most cases, it would go back to the same judge. But obviously with a different judge, you might get a different sentence, but at least that would be based on acceptable evidence in a sentencing proceeding, as opposed to this unduly biased emotional evidence.

Dave Bittner: All right. Well, we'll have a link to that story in the show notes. Let's take a quick break here to hear from our show sponsors. We'll be right back after this message. [ Music ]

Dave Bittner: Ben, I recently had the pleasure of speaking with Garfield Jones. He is Senior Vice President of Research and Strategy at QuSecure, and also a former CISA Homeland Security Executive for Quantum. And our discussion centers around a recent OMB memorandum on migrating to post-quantum cryptography. Here's my conversation with Garfield Jones.

Garfield Jones: So, this White House executive order is one of those that's really a call to action. And it's really more of an, almost an action memo than a actual high-level, than the other executive orders. The other executive orders did provide some action, but this one really shifts from the beginning phase of PQC, or post-quantum cryptography, to more the implementation phase of it. So, knowing that we have to transition, they were looking for more of an execution instead of just discovery from the first White House executive order to now the more I'd say robust executive order, which we're looking for more than discovery. We're looking for implementation. We're looking for, with better dates. They've also moved the dates from 2035 to 2030 and 2031 for key establishment and digital signature respectively. So, they're looking for, they're trying to establish more of an action plan of moving towards post-quantum cryptography.

Dave Bittner: So, for the organizations that do business with the federal government, what's the message here?

Garfield Jones: The message is get your button gear [laughter]. So, you know, we've only got a few years left here on this transition. And I mean, honestly. if you haven't started the transition, you're way behind. At first it was 2035. And, like I was saying earlier, so we had some runway, but, you know, a four-year shift left to, you know, shortening the timeline. Organizations have to really, that are doing business with the federal government, have to really start moving. I mean, you're looking at documents. The document CNSA 2.0 was probably the one that really told organizations, if you don't start moving, we are not going to use you in 2027. CNSA put some really early dates in there, and I don't think organizations that were doing, that want to do business with the federal government really paid attention. CNSA said procurement for non-PQC items basically were, they were going to stop at the end of 2027 or, yeah, somewhere around there. And I think organizations were not, did not take that seriously. And that was a precursor to what came down here, is now we're looking at changes to the FAR, the Federal Acquisition Regulation, which is basically, you know, we all live by in the federal government. Especially if you want to do work with the federal government, they're going to make changes to that document, and that's going to impact a lot of organizations or businesses that want to do work with the federal government. It's going to, you know, say, if you're not PQC enabled or PQC ready, or you're not transitioning to PQC algorithms, basically cannot do work with the federal government. And that's going to be really impactful for those.

Dave Bittner: And do we have a high degree of confidence that those restrictions are going to hold?

Garfield Jones: Oh, absolutely. Like I said, you know, there were other documents that were put out before that kind of gave a, the storm clouds were coming, right? The storm clouds were there. And, you know, it's up to you if you want to, you know, sit out in the rain. But they're definitely going to hold because the federal government is, has to protect its data. It's got a lot of sensitive data. We can't fool around with this. It's, you know, the sensitive data is going to be, you know, we've already, we're already behind the curve on, beyond the curve on a lot of the data that we have today with the harvest now, decrypt later problem. And the federal government knows that this is a threat, and they are taking steps to mitigate that threat. And so, you know, these, I think, honestly, I believe that they're actually going to start tightening even more. They're probably going to start, you know, really putting the squeeze on a lot of things that we are normally accustomed to. I mean, you, we've seen some things in some of the other documentation that's following the executive order, that there's a lot more, there's a lot tighter restrictions on things that are very commonplace in the, yeah, basically in the businesses that they use. So I think you really have to be careful that you don't take this too lightly and you start moving. These restrictions will hold.

Dave Bittner: What's the response from industry been to the timelines that the Office of Management and Budget have put out here? Are they being perceived as being realistic?

Garfield Jones: To a certain extent. They are being realistic. I think that you have to be, the 2030 and 2031 timelines are realistic for organizations, but they have to have already kind of started moving. If you're talking about agencies in the federal government, if they're smaller, they might have time to move over. If they're larger, it's going to take a longer time and it's going to be a lot more difficult and they may not make it. If we're talking about private organizations, just the industry as a whole, the response has been, I mean, has been very, you know, trying to move to it. They're starting to implement TLS 1.3 in certain areas. You know, I was just at Quantum Tech, one of the conferences. And the response there was odd because that same week the EO dropped and we were like, oh my goodness, you know, and the uptick in just how people were reacting and the, you know, the reaction to the EO was just amazing. And I think, you know, we were like, wow, all right, finally, we've got deadlines. We've got, you know, NIST, you know, OMB, everyone's really starting to point to the same directions. You know, NIST has always been working on this since, like, 2016. OMB actually, their memo actually, you know, says the same timelines as the executive order, but their full migration doesn't end until 2035. So we've got the HVAs that they're kind of starting with. I mean, I agree to a certain extent with HVAs. But you've got a lot of interdependencies that you have to kind of worry about. And so HVAs are great to start with, but you've got to worry about all the assets that are connected to them. So, it's not going to be as easy as just prioritizing HVAs and moving out. So, I think a lot of the industry has been very responsive in a very positive way to this executive order because they're like, yes, we have the firepower now to put in front of our senior management to say, here, this is, we've got to move.

Dave Bittner: Right.

Garfield Jones: The president is saying this, you know, we can't sit on our hands and be like, oh, you know, next year, five years from now. So we've got to move now.

Dave Bittner: That's funny you say that, because, you know, one of the, I guess for folks who've been tracking this for a long time, you know, one of the jokes about quantum encryption for a long time was that it was kind of like fusion energy, you know? That it was always a few years off, no matter when you ask, right?

Garfield Jones: Exactly.

Dave Bittner: But here we are, as you say, it's a good thing that we finally have some solidity in our deadline.

Garfield Jones: Yeah, absolutely. I mean, these executive orders are always, I'm always in favor of executive orders because I think anytime you get direction from the highest office in the land, people are going to take it seriously. They're not going to be like, oh, you know, we're going to do it next week. It's going to be taken seriously. And I think businesses already know that, okay, well, if the federal government is moving towards this, we've got to beat them there because we've got to be able to work with them. We've got to be able to interoperate with their assets. We've got to be able to be included in that FAR document so that we can still have that business continuity. It can't be taken lightly. So I think it's great that we have this kind of power and that those two documents, the two EOs that dropped, the OMB memo that dropped, you know, that's, those are things that are really needed to kind of light a fire on their businesses and senior management.

Dave Bittner: Well, given these realities, what are your recommendations for organizations who are in this space? What sort of things should they be doing?

Garfield Jones: They should definitely be, I've always talked about the awareness and preparedness and action part of this. And I think awareness is always important. So they should start, you know, educating their folks, their resources, their human resources. Start following some of the guidance that the executive order about appointing committees and people to kind of lead this transition, if they haven't done it already. If they have, then now you need to move to the other part of the executive order, where we're looking at, you know, what assets are most important to you? How do we discover those assets, prioritizing those assets? Implementing, one thing, I don't think it was really discussed too much in the executive order was the funding. You need to start allotting the funding for this transition. It is not going to be cheap. There's assets that we haven't used in a while that, you know, it's like, you know, I'm a little older now and I tried to go out there and play soccer and, you know, things don't always work the same way when you haven't used them in a while.

Dave Bittner: [Laughter] I'm with you.

Garfield Jones: So you've got to be able to find the assets that you haven't, you know, used in a while and be able to account for those and be able to, you know, and prioritize those. So, you have your HVAs, you have your operational technology assets, you have things that you, you know, that you need to prioritize to moving towards the PQC. You also have TLS 1.3. TLS 1.3 I always say is the low-hanging fruit. If you can move to TLS, if you can transition to TLS 1.3, that's already a win, right? You're already protecting yourself against that harvest now, decrypt later. So there's threats out now. So harvest now, decrypt later, we want to be able to make sure. So going into TLS 1.3. The other piece of it is try to, you know, talk to vendors. And talk to vendors and people who are, or businesses who are helping transition organizations towards PQC. Look at how to implement crypto agility. And the reason I mentioned that is because, you know, as we move towards this transition, we have three algorithms that are set right now. We have the FIPS 203, 204, and 205. And that's M-CHEM, ML-DSA, which is digital signature. And then we have the SLH-DSA which is the, another one for digital signature. So we have one key establishment and two digital signature algorithms. And I think talking to the organizations, helping them, having them move to, you know, implement crypto agility, then you're ready for the next set of algorithms that come out. So if those don't, if those happen to be deprecated or we have problems with those as we move through, then we have, then we're ready for a new algorithm. So, you know, the awareness part, the education, being able to move to that aspect of it, you know, the discovery, you know, prioritizing your assets and everything like that. The crypto agility is another piece of it. And of course we have to have some kind of reporting mechanism for it so that we know what we've done, what we have to still do, and so forth. But you really have to make sure you start with those elements that are important. And I didn't mention the last part to summarize was the funding part. That's the most important part. A lot of funding for this transition. As I said, it's not going to be cheap. And the longer you wait, the more expensive it gets. It's like, you know, free agency in the NFL.

Dave Bittner: I guess it's, as you mentioned, it's easier, or bends things in the security professional's favor if they can walk into the boardroom with an executive order rather than, you know, a memo from a low-level bureaucrat somewhere, you know, buried in a government organization, right?

Garfield Jones: Exactly. Yeah. You know, tech documents are great. You know, we can, you know, I read those articles on the, you know, on the academic databases, you read them in the Nature magazine and everything else. No, it's coming, I promise you, you know? Yes, sure, okay. But when you have an executive order, even though it's only a few pages long, you know, it's going to weigh heavy on that desk. It's going to be, you know, you're going to put it down and say, I have it right here. And do you need something else on this sandwich here, let me go ahead and drop this OMB memo as well on top of it, you know.

Dave Bittner: Right.

Garfield Jones: Let's put some lettuce and tomatoes in between this bread that we just dropped here. So, I think once you have those kind of, you know, meaningful dates and documents, that you're good to go in terms of being able to help your organization move either towards the cloud, you know, with on-prem assets or, you know, with Edge assets in the Edge cases as well. So I think you have all the ammunition you need right now. You just need to execute.

Dave Bittner: Ben, what do you think?

Ben Yelin: I still understand very little about quantum computing in the first place.

Dave Bittner: You're not alone, my friend.

Ben Yelin: And post-quantum cryptography. I feel like there should be a national freely available course on quantum.

Dave Bittner: Yeah.

Ben Yelin: I actually assume there are courses out there.

Dave Bittner: There are, but it's one of those things, you know, I think it was like Richard Feynman who was one of the folks who originally came up with many of the things that we understand now about quantum physics and all that sort of thing is they just talk about how unintuitive it all is.

Ben Yelin: Right.

Dave Bittner: And for most people, if they claim to understand it, they're probably mistaken.

Ben Yelin: It's just very difficult, for me at least, to conceptualize it.

Dave Bittner: Yeah, I'm with you.

Ben Yelin: So.

Dave Bittner: It's one of those things where I just nod my head and trust that very smart people --

Ben Yelin: Hope other people know what they're talking about.

Dave Bittner: Right, very smart people are taking care of this for us. But obviously Garfield Jones is one of those very smart people. And we do appreciate him taking the time for us. [ Music ] And that is "Caveat", brought to you by N2K CyberWire. We'd love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like our show, please share a rating and review in your favorite podcast app. Please also fill out the survey in the show notes or send an email to caveat@n2k.com. This episode is produced by Liz Stokes. Our executive producer is Jennifer Eiben. The show is mixed by Tre Hester. Peter Kilpe is our publisher. I'm Dave Bittner.

Ben Yelin: And I'm Ben Yelin.

Dave Bittner: Thanks for listening.