Cyberwarfare in the AI Age: A Conversation With Jen Easterly

24:18Jen Easterly:There's no evidence that a malicious actor changed, altered, or deleted a vote or had any impact on election security. But look, I am a nonpolitical person. I'm a lifelong independent. I'm somebody who's focused on security and cybersecurity has to be, and certainly this is the way I treated it, a nonpolitical, nonpartisan endeavor. And even as we had, um, the, the, what's called the lead government role for election infrastructure security, we had fantastic relationships with, you know, red States, blue States working with the secretaries of state, because the secretary of the state who are the chief election officials in most cases, nobody wants to run a fraudulent election.

25:00Jen Easterly:Everybody wants to have the technical resources that we provided, both cyber and physical, to make sure that their infrastructure was safe and secure. So I was very proud on the morning after the presidential election in November of 2024 to put out the statement that said, you know, this election was safe and secure. And it wasn't because, Dan, it was quote unquote, too big to rig. It was because the infrastructure was in fact not compromised. We had no evidence of any compromise that had any type of an impact on the election. Look, it's really important to understand that the infrastructure that we use, the systems that we use to vote, they're not connected to the internet.

25:45Jen Easterly:It's very hard to hack something that's not connected to the internet. It's one of the reasons we need to be very thoughtful about some of these new AI systems and really make sure that they are, in fact, contained. And that's why the guardrails that OpenAI is going back to look on to make sure that these systems cannot do things like jump into completely air-gapped, non-internet-connected systems. It's sort of one. So very hard to hack something that is completely disconnected. Two, I think the number is probably almost 98 % of voting records are paper. Very hard to hack paper, right? And three, there are multiple, multiple layers of cyber and physical controls in place to protect election infrastructure.

26:29Jen Easterly:So even if there is a vulnerability, which there are, there's vulnerabilities in all systems. There are multiple layers to protect that vulnerability from being hacked, exploited, or weaponized. The last thing is, you know, if you talk to any election official, if you've seen one state's election, you've seen one state's election. Every jurisdiction does it a little differently. And that actually creates incredible resilience because you don't have all of these things connected. So it's very, very hard to do anything that lands at scale, which is why I can say with fidelity that, in fact, those elections that I certainly observed or was part of was, in fact, secure.

27:11How worried are you as we head towards the midterms in a few months and then obviously the 2020 presidential election? You wrote a piece when you were head of CISA, which addressed both cyber risk that you were just talking about, as well as kind of AI enabled this information. The capabilities a few years ago when you wrote that look like nothing compared to where they are now. How bad are those threats as we head towards these elections?

27:33Jen Easterly:If I'm remembering that piece, the TLDR is basically like, we don't need to be that concerned about AI having any impact on the 2024 election, but we will likely need to be concerned about future elections. And to your point, Dan, these capabilities have advanced so fast in such a short period of time. We should be very thoughtful to the question that you're asking now for the reasons that I talked about the resilience of election of election infrastructure The fact that it's not connected to the internet the fact that you have paper the multiple levels of physical and technical controls All of that still holds frankly What I think is The most worrying is not actual technical hacks But perception hacks because we live in an environment, an information environment that is so fragmented that it doesn't necessarily matter whether something is true.

28:31Jen Easterly:It matters whether it goes viral. And so I think at the end of the day, the people who have to actually certify those votes, those chief election officials whose responsibility it is to run safe and secure elections, I think they're going to do everything they can to make sure that their voters have confidence. And so there will be, I think, a pretty messy information environment out there. But we'll have to look towards the states because it is by the Constitution their job, not the federal government's job, to run elections. I have great faith and confidence having worked with chief election officials across the country.

29:15Jen Easterly:So it will be up to them and the governors to make sure that when those elections happen, and I believe they will be safe and secure, that the final messages that come out about who won those elections, that voters can have fidelity in those results and can believe that their votes were counted as cast. We'll return to my conversation with Jen Easterly after a short break.

29:47What if you could explore places in the news like a reporter does? I'm Nicholas Wood, a former journalist with the New York Times and BBC. And 16 years ago, I created the travel company Political Tours. Our small groups are led by top correspondents around the world. In the next few months, we're off to Mexico, followed by South Africa, Japan and the French presidential elections. Come and join us. Go to politicaltours.com. That's politicaltours.com.

30:42Jen Easterly:To learn more, visit independenthq.com. Did you know that Foreign Affairs editor Dan Kurtzvallin sends a free newsletter every Saturday? Visit foreignaffairs.com slash spotlight to get the editor's spotlight. A weekly email from Dan featuring more of the helpful context and clear analysis that you get here on the podcast. That's foreignaffairs.com slash spotlight. Sign up today.

31:11You talked earlier about the change in Chinese doctrine that you witnessed and came to understand grappling with the salt typhoon attacks a few years ago. As you look at Chinese AI-enabled cyber capabilities, how do they compare to ours? I mean, there are obviously Chinese models that are perhaps a few months behind Mythos, but how do you assess those capabilities? and what should we be doing to prepare for the threat from Chinese models, which we have obviously less ability to control?

31:40Jen Easterly:Yeah, I think this is another, you know, again, if I wasn't so worried about everything, I would just like, it's also so exciting and so fascinating, right? I mean, I often say like, this is the most exciting time to be alive and to be in my field. But you know, the stakes are obviously really, really high as well. And so one of the things that happened with this Hugging Face open AI incident. I don't want to nerd out on you too much, but when Hugging Face realized that they had been, quote unquote, breached, they tried to use a model. I think they were trying to use Anthropic to do what's called the IR, the incident response, to help them identify what happened and to clean up the situation.

32:24Jen Easterly:And the model they tried to use essentially thought that they were malicious, a malicious actor, and shut them down from using it. And so what did they do? They went to use a Chinese open weights model, GLM 5.2 that did a good job. And open weights means that you can manipulate it more than you could with a closed. Well, yeah, you can download the model and use it, right? And you can see that the Chinese open weights model, again, these are models that can be downloaded and they can be modified. That's the security issues around them, but they're much cheaper, right? So you can download them. and use them.

33:02And frankly, they're becoming better and better and more capable.

33:06Jen Easterly:So the latest models I mentioned, GLM 5.2, I think that comes from Z.ai, then Moonshot AI, you'll, you know, people will probably have heard of something called Kimi K3. There are all these Kimi versions. Look, if you don't have the money to spend on some of these more expensive models, you may want to access some of these Chinese models. Now you might be saying, Jen, you're crazy. Like, aren't you telling us about all these threats from China? And yes, I think it's a legitimate concern, but you can download these models. So you're not connecting in any way to China. So, you know, are there concerns about, you know, has China put some sort of a capability like a backdoor or, you know, could China somehow put a control in there that can affect how you're able to use this capability.

33:58Jen Easterly:So I think just as you would want to rigorously test anything that you didn't completely trust, yes, you would want to do that. But in many cases, if they're cheaper, if you don't have a huge risk in terms of how you're using them, well, you may want to tap into those models. And so this is this huge issue that's going on right now, whether in fact, you know, the US government, there's been a little bit of chatter about will the US government somehow put controls preventing access to Chinese open weights models. There was a letter signed, I think, by NVIDIA and many of the big companies saying we need to continue to have access to open weights models and how important they are to have both an open and closed system.

34:45Jen Easterly:And I strongly believe we need to have access to open weights models. You know, there may be very good reason to tap into certain types of the closed models from OpenAI and Anthropic, but I think a thriving technology ecosystem is one in which there is greater competition. And that competition will hopefully push people to ensure that these models are not only very capable to task, but that they are as secure as possible. And, you know, if you look at how China is approaching these models versus kind of how the U.S. in a more ad hoc way in terms of how you do the policy, right, they put in place the executive order that had voluntary testing.

35:27Jen Easterly:There's some talk about I think they're going to release a framework for this testing. There's a vulnerability gold eagle capability that Treasury is running to bring together critical infrastructure to identify and remediate vulnerabilities. But look at what China is doing if folks are following. They just had this World AI Conference that just concluded. And there were a couple really interesting things from it that were very specifically focused on AI safety. president Xi was there for the first time he attended in person and he gave a speech that was particularly like safety forward and they did this official conference statement that that was very much focused around cyber security risks and then they just released this information about I guess they call it Waco the world AI cooperation organization the first intergovernmental body dedicated to AI.

36:26Jen Easterly:But the fact that you had President Xi specifically talking about the importance of safety and security, the worry about loss of control, the worry about cybersecurity capabilities, I actually think we should sort of pause on that and recognize that we want to make sure that our population is safe and secure. You know, even though China has greater threshold for pain. I think ultimately they want to make sure that their population, their economy, their place is secure. And so this is why I think one of the most important things that has to happen over the next six months is the US and China have to come together and have very serious discussions about how to ensure the safety and security of these increasingly powerful tools, not just because of the cyber implications, but because of the implications for bioweapons or chemical weapons, the full range of concerns around that, the loss of control.

37:29Jen Easterly:Look, we were able to do this in the nuclear world. It's a little bit different now because, of course, nuclear weapons were built and safeguarded by governments that were disincentivized to use them. These capabilities are largely built by private companies that are incentivized for returns to their investors and profit motives, but it is the responsibility of governments to protect their people. And so we have to have these discussions and we have to figure out how to put the right guardrails around these capabilities so that these very, very powerful, magical tools are able to serve humanity in powerful and positive ways and not do irreparable harm.

38:11Why should we think it's possible for those serious discussions to happen. I think there has been some agreement on at least starting to have them in both the Biden-She summit and the more recent Trump-She summits. But the reports on those do not suggest that they've really gotten to the heart of the matter or grappled with the toughest issues. Do you think that there is, in fact, scope given the kind of shared existential risks, despite the competitive dynamics, to really get something done?

38:37Jen Easterly:Look, hope is not a strategy, but we have been able to do hard things, right? In the Obama administration, there was an agreement that was done, as you well know, with the Iranians on the nuclear issues, right? So we've done hard things before. We've been able to put a nuclear test regime in place, controlling AI, putting safeguards in place, having testing regimes. None of this is against the laws of physics. These are not impossible things. These are a matter of incentives and will. And I think from reading in the press, there is moves afoot led by the Treasury Secretary to have these discussions about AI safety in the lead up to the summit, the upcoming Xi summit, if it in fact happens in September.

39:33Jen Easterly:I was very pleased to see that. And I can only imagine that this incident of the past week will further sort of motivate these types of discussions. So could it all fall wildly apart? 100%, right? Could we start to edge towards a place where we make sure that these capabilities protect our population? We certainly got to try. How do you think about other actors beyond the U.S. and China? I think some of the concern here is that while China is, by the standards of some of its other partners, relatively has low risk tolerance, wants stability, cares about the worries about the proliferation of some of these tools, that probably isn't true to the same extent with Russia or with North Korea, let alone with terrorist groups or financial criminals or anything else.

40:29how do you think about that dimension of it, given how easy, especially with open-weight models, as you noted, it is for some of those actors to turn these into pretty powerful weapons in a way that you just could not with nukes?

40:40Jen Easterly:Yeah, I mean, I think that's exactly the question. That's the incentive, right? The U.S. doesn't want, China doesn't want these capabilities falling in the hands of, let's just set aside the other nation states for a sec, falling in the hands of cyber criminals, you know, worst case terrorists. I know we don't spend a lot of time talking about global terrorism. Having served for 21 years in uniform, having been in Iraq twice and Afghanistan and, you know, served much of that time after 9-11, I am very, very mindful of a world where terrorists can get a hold of these types of capabilities and wreak enormous havoc.

41:22You know,

41:23Jen Easterly:One of the stories that I tell that always sticks in the back of my mind, I work for Condi Rice after 9-11. I was there in the White House from 2002 to 2004. One of the last things I did before I went off to the National Security Agency at Fort Meade, I went with her. She testified before the 9-11 Commission. You remember, that's one of the best books out there on national security, terrorism, and threats. And, you know, the big takeaway from that book, from the hearings, I think, was, you know, the famous quote, which is, on that September day, we were unprepared. We did not grasp the magnitude of a threat that had been growing for some considerable period of time, a failure of policing, a failure of management, a failure of capability, but above all, a failure of imagination.

42:11right and here we are almost 25 years from 9-11 and the velocity of information and the velocity

42:20Jen Easterly:of technology are just coming at us so quickly that i worry you know the the greatest danger is not the failure of technology it's the failure to imagine how this technology can be weaponized so you know the second quick story i'll tell is when i was the commander of the army cyber battalion And we had gotten this information that Al Qaeda was going to publish its first English language magazine called Inspire. Very slick little magazine. You might remember it was from the 2010 timeframe. Remember that? And it was, there was a page in this magazine. It was called How to Make a Bomb in the Kitchen of Your Mom.

43:02Jen Easterly:And it was literally a step-by-step recipe of instructions on how to build a homemade bomb. When I was in Iraq, we dealt with these things called improvised explosive devices from, you know, Iraq and then even Iran back then, these explosively formed penetrators having, you know, thousands of killing thousands and thousands of Iraqi civilians in our troops. But we were very worried that this was going to then lead to lone wolf terrorist attacks in the West. And we tried to figure out, was there a way we could prevent this thing from getting out? And we really, really worried about it, worried about it.

43:36Jen Easterly:And ultimately, there was nothing we could do. It got out. And we really held our breath, Stan. It was literally like weeks or hours and days and weeks and months. And we kind of thought we dodged a bullet. But then three years later, two brothers in Boston, Tamra Lynn and Joe Karzarnaya, used that exact same Inspire magazine, that exact same recipe, how to make a bomb in the kitchen of your mom to build the pressure cooker bombs they left at the Boston Marathon. and that killed of course three people including an eight-year-old boy and maimed and wounded hundreds and hundreds of others and my point is just think about like that's so simple right a page in a magazine step-by-step instructions now think about what ai can do think about that ai generated recipes precision imagery auto-generated shopping list now available when people can say well you can go to the library or you can research this anybody that works with these tools knows how easy it is to be able to access information.

44:40Jen Easterly:Just imagine being able to access this type of information to create cyber weapons, chemical weapons, even crude versions of bioweapons. Like that's something we don't wanna think about because it's a horrible, frightening thing. But at the end of the day, we have to be able to imagine and embrace both the best of what's possible, but the worst of what's possible to make sure that we're planning for it, We're preparing for it so we can respond, we can recover, and frankly, so we can build the necessary resilience to be able to protect the American people on the worst case scenario. But we need to do everything we can to get ahead of that worst case scenario.

45:22Jen Easterly:And, you know, we have maybe months, Dan, to get ahead of that. And that's why I think it's so important to have these conversations, both across the business community, but to ensure that the business community and the technology community are connected in to governments who are there to be able to put the right guardrails in place to protect their people. It was striking how much the Trump administration's policy on AI changed as soon as Mythos was released, going from the kind of deregulate everything and let it rip, I think was the phrase that someone in the administration used, to something that looks, at least to my inexpert eyes, not that different from what the Biden administration would have been doing.

46:05Do you think that U.S. policy is at least directionally where it should be at this point?

46:11Jen Easterly:um you know i i understand the enthusiasm why folks would say well let's you know let's go go go on on this technology i think there's a bit of a failure of imagination in terms of you know only thinking about the incredible magical things they can do but not being willing to think about all of the downsides i think there's a little bit of the idea that that we've had frankly for for years and years. I mean, this is really kind of the cultural ethos that have brought us so many good things, but moving fast and breaking things has also resulted in a lot of breakage, Dan, right? It's why you have an internet full of malware, why you have software full of vulnerabilities, why you have social media full of disinformation.

46:54Jen Easterly:And as been discovered, now we are racing, and that's the word that's used, right? Racing to build and operationalize the most consequential technology of our lifetime. And so I think now that we can actually imagine what the worst is when Mythos came out, followed on by ChatGPT, 5.5 Cyber, and now some of the Chinese models that are getting equally capable, I think the government has recognized some of the damage that could be done. First and foremost, it was really about the global financial ecosystem, system which can not only be impacted technically but also from a confidence level perspective right and so that was I think the first wake-up call that led to the executive order of early June now there's a word in that executive order which is voluntary and I understand that there is this tension between well if you start putting significant safety regulations in place that could then crush the innovative spirit that really makes American technology what it is.

48:02Jen Easterly:And so I understand the trade-off that people are grappling with. But at the end of the day, I think the government's fundamental job is to protect the American people. And being willing to be very open, again, to the good, but also the bad. I think the bad leaves you in a place where you can't really do much in a sustainable way through executive orders. So the administration is taking certain steps. Should they take more fulsome steps that are not necessarily voluntary? You know, you could argue that the Biden administration may be over-rotated by trying to find this novel Defense Production Act capability in their 80-page executive order.

48:46Jen Easterly:You know, you could argue like both something in the middle might be a little bit better. But Dan, And at the end of the day, executive orders only go so far. You can get rid of the whole thing when a new president comes in. What we need is the United States Congress to legislate. They need to create legislation that will actually put guardrails around the most dangerous capabilities. And again, this is not against the law of physics. This is actually happening at the state level. If you look at California, you look at New York, you look at Illinois' latest legislation, they're actually, it's hard to get right, right?

49:25Jen Easterly:It's really hard to get this right. But they're pretty good in terms of a roadmap, a framework for how at the national, at the federal level, you could look at creating something that would, again, enable us to leverage these capabilities in very powerful and positive ways, but mitigate their most serious and damaging harms to the American people. And so the administration is doing what the administration is going to do. What we need is the United States Congress to actively legislate. slate. One more worst case scenario I want to make sure we get your thoughts on. Wow, we haven't talked about enough worst case scenarios.

50:02Jen Easterly:That's right. One more, one more. Quantum computing. There's this fear that once quantum reaches a point, all of a sudden, all of this looks much worse. When I first heard the phrase, harvest now, decrypt later, the idea that the Chinese government and others might be just sweeping up all of our files, which they can't decrypt now, but someday they'll be able to once quantum capabilities reach a certain point is a chilling thought. How do you assess that threat and what should we be doing to prepare for it? Yeah. I mean, I think that part of the good news here is this is not an intractable problem.

50:39Jen Easterly:We did a lot of work on this when I was at CISA along with the technical experts at NIST in the commerce department and then at the NSA. And they work with the private sector to create create algorithms that were quantum safe right so there is actually a way to safeguard your infrastructure put quantum safe encryption in place to safeguard against what's called a cryptographically relevant quantum computer right the quantum computer where if you can think about the matrix where all the the secrets are all revealed and in traditional encryption is all broken. Will that necessarily solve the harvest and decrypt issue that you just pointed out?

51:21Jen Easterly:No, but it will allow us going forward to protect the safety and security of the current infrastructure. So one of the most important things that critical infrastructure entities need to do now is to start making that transition to quantum safe encryption. And again, these algorithms exist. The issue is like it takes time and resources. And so you have to have a roadmap and plan. You have to identify those crown jewels and then you have to start the transition. But again, it's not an intractable problem. It's just a time consuming and expensive problem. But it's one of the most important things that big critical infrastructure companies need to do.

52:05Jen Easterly:And the good news is like people are asking that question. Two, three years ago, I don't think anybody was asking that question, Dan. And so companies recognize what they need to do and they're starting to put in place. I think Microsoft just came out and noted that, you know, that the move towards quantum safe encryption, they're actually calling 2029 of when they need to transition. And so, you know, that's less than five years from now. So we really need to look at a compressed timeline and, you know, anybody, again, you know, in this administration, there's been a lot of focus on post-quantum cryptography.

52:41Jen Easterly:I think it's one of these issues that, again, it's not a political issue. It's not a partisan issue. It's a technical issue of protecting the American people and the critical infrastructure we rely upon every hour of every day. So some still really good work going on at NSA and at NIST if folks want to take a look at that as a roadmap for transition. This may fall into the category of good news. I'm not totally sure. you spent time when you were in the army helping set up cyber command. You worked on cyber issues and cryptography issues in Iraq and Afghanistan. We don't often talk about our own offensive capabilities.

53:17I assume those are as good or better than any we are worried about coming from other governments or militaries. How does offense fit into this picture? What role should it have? Do you see kind of evolution that should be underway right now when it comes to our own offensive capabilities, whether that's from the military or NSA or other agencies?

53:39Jen Easterly:Yeah, it's a great question. Thanks for asking. I've been lucky and privileged to have so many great opportunities throughout a career that has spanned a bunch of different things. But one of the funnest and coolest and most rewarding and most challenging was working with a small handful of people to set up US Cyber Command. This was in the 2008-2009 type frame in the wake of something that was called Operation Buckshot Yankee, which was Russia going after classified military networks. It was a watershed moment because it was another one of those like, why didn't we think about this? and that then led to secretary gates at the time telling my boss general alexander who was the director of nsa we need to actually set up a capability both to defend military networks from significant nation-state threats as well as to be able to project power in cyberspace that offensive uh cyber and that's one of the reasons why very uniquely cyber command the commander of Cyber Command is dual-hatted as the director of NSA because the idea was is you needed to set up this capability on top of the cryptologic enterprise because understanding those vulnerabilities, right, the foreign intelligence aspect of how do I identify those flaws and defects in our adversaries gives me the ability to then take advantage of those adversaries for offensive operations.

55:09And, you know, we set up in 2009 or I guess 2010 final operational capability

55:15Jen Easterly:and it evolved a little bit in fits and starts, but I think it's a really good new story to look at the capability and capacity of Cyber Command now. And I've been really, really proud to see the talent, the resources that they've been able to bring to bear. You've heard about things in Iran heard about things in Venezuela. They're things I'm sure that we haven't heard about. But it's great to see the capability flourish in a way that we envisioned and to be another tool in the toolbox for protecting the American people. Is our offensive doctrine similar to China's at this point? Do we think about those in similar ways?

55:53Is there a certain kind of balance there? I think the one distinction that we could probably make between the U.S.

56:00Jen Easterly:and China is we typically will look at civilian infrastructure different from military infrastructure. And I think China does not take that view, which is why when we talked about those cyber bombs in our water systems, our power systems, our transportation and communication systems, I don't think China really focuses on whether they serve civilian populations or military populations. I think that is one major thing that we've looked at differently because of all the human rights and the laws of war implications. Now, at the end of the day, if China is willing to hold our most important critical infrastructure at risk, I think we need to be very thoughtful of all of the things we need to do.

56:42Jen Easterly:cyber being one of them, right? Cyber is a tool. It's part of the domains of warfare. But I think probably this administration is relooking how we need to approach holding our adversaries at risk in the same way that they're holding us at risk. Well, Jen, I feel like we'll have to maybe do another one of these in three or six months when we see how well or not the United States and others have done preparing for all of the threats that you have helped us imagine in full color. I appreciate that and appreciate the fact that you made it all so clear. So thank you so much. Awesome. Thanks so much.

57:16Jen Easterly:So great to see you. Appreciate the combo.

57:24Thank you for listening. You can find the articles that we discussed on today's show at foreignaffairs.com. This episode of the Foreign Affairs Interview was produced by Ben Metzner and Kanesh Kleror. Our audio engineer is Todd Yeager. Original music is by Robin Hilton. Special thanks as well to Arena Hogan. Make sure you subscribe to the show wherever you listen to podcasts. And if you like what you heard, please take a minute to rate and review it. We release a new show every Thursday. Thanks again for tuning in.

58:12Jen Easterly:We hope you enjoyed this episode of the Foreign Affairs Interview. Don't forget to visit foreignaffairs.com slash spotlight to sign up for Dan Kurtz-Falen's free weekly newsletter featuring more of the clear-headed analysis that you enjoy here on the podcast. Sign up today at foreignaffairs.com slash spotlight.