Jen Easterly and Sami Khoury, the top U.S. and Canadian cybersecurity officials, respectively, have lived almost parallel lives over the past year. They both assumed their current roles within a month of each other in 2021, heading relatively new government agencies that were both created in 2018. As neighbors and allies, they work closely together and share the same mission, against the same adversaries.

That mission has become more critical, and those adversaries have gotten bolder. Russia’s cyberwar in Ukraine continues in parallel to its missiles and munitions, and China’s hackers have reminded the world of their prowess in recent months by infiltrating the emails of senior U.S. government officials and infecting military bases with malware. Artificial intelligence, meanwhile, has implications for offense as well as defense in cyberspace, but its exponential growth makes it a major challenge on both fronts.

Easterly, the director of the U.S. Cybersecurity and Infrastructure Security Agency (CISA), and Khoury, the head of the Canadian Centre for Cyber Security, sat down for a joint interview with Foreign Policy last week on the sidelines of the Billington Cybersecurity Summit in Washington.

This interview has been edited for length and clarity.

Foreign Policy: What’s the biggest change you’ve seen in the last two years?

Sami Khoury: So I started in August 2021, and I knew that cyber was a big deal and we were dealing with it, but I think what changed in the last two years is how bold some of the cyberattacks are getting or some of the cyberactors are getting—but also on the flip side, how seized partners are to the threat of cyber. So the message is starting to resonate.

Jen Easterly: I think certainly our partnership with Sami’s team in Canada is incredibly close—as well as with the other Five Eyes [countries]—and just the work that we’ve been able to do has been pretty extraordinary. I think the transformation that we’ve been able to catalyze with our partners across the globe to secure the global cyber-ecosystem is much more advanced than I would have expected. Much more work to do, but I look back over the past two years, and I am optimistic.

FP: How much of a catalyst has the Russia-Ukraine conflict been for multilateral cyberpartnerships?

JE: We have been working very closely with [Ukraine] on a variety of things to help improve their cyberdefenses, but just as much, at least from a CISA perspective, we’ve been learning from them about cyber-resilience and more broadly about operational resilience. We’re looking at the next phase of that, which is really how do we anticipate significant attacks—we’re particularly concerned about China—and anticipate disruptions so that we can operate through disruptions to enable recovery, remediation, and drive down risk.

SK: We started warning Canadian businesses as early as possible to be vigilant, to be on top of their IT, to report any incident small or big. We were definitely concerned about spillover into Canada. We know Russia is a formidable adversary and has cybercapabilities that they have demonstrated indiscriminately and irresponsibly. We share two of the most connected countries in the world, Canada and the United States. On energy, electricity, natural gas pipelines, finance—that connectivity north to south, there is no equivalent elsewhere.

FP: Why do you think we haven’t seen a major Russian cyberattack in the West since the Ukraine invasion, and what does that say about Russia’s abilities?

JE: Russia—Russian President Vladimir Putin specifically—miscalculated on some really big things. I think they thought that the march to Kyiv was going to be done in a week and they would take over the country, that they could quickly overmatch Ukraine’s capabilities with their military capabilities, that the United States and the global community would not be as united as we are in terms of supporting Ukraine with materiel, with funding, and the moral support that comes with all of that, standing up to a bully like Russia.

Where I think they didn’t miscalculate is based on the warning that U.S. President Joe Biden gave to Putin following the Colonial Pipeline attack [in May 2021], where he said: Do not go after our critical infrastructure. I think Putin correctly calculated that that would be overly escalatory, and so I think that’s a bit of deterrence by punishment. But I also think that there’s a bit of deterrence by denial. We saw a significant effort to harden networks, to make changes and security enhancements.

I don’t think we should take the lesson that Russia has no cybercapability. We know that they do. They’ve been very focused on Ukraine, not focused on the rest of the world, but I don’t think we should assume that we will not suffer any more attacks. All that said, I am now more concerned about the formidable threat from China.

FP: How does China operate differently from Russia in cyberspace?

SK: We named Russia and China as our two top cyberadversaries. I think China is a strategic threat—it’s a long enduring threat. Maybe Russia is more tactical in some cases, but definitely we are worried about the scale and scope of Chinese cyberoperations. Everything from spying on the government to pre-positioning your critical infrastructure to asserting space and technology development, disinformation campaigns, and so on. So it’s a multifaceted threat that they pose. They don’t share the same values that we do. They are active on many, many fronts, and we need to tackle them all.

JE: I would point people to the annual threat assessment that our intelligence community released earlier this year, because it’s a pretty stark warning. And it talks about specifically—just looking at the geopolitical landscape and what you’re seeing from Chinese leadership—in the event of a conflict, China will almost certainly consider aggressive cyberoperations against critical infrastructure and almost certainly has the capability to disrupt pipelines, to disrupt rail, and other critical infrastructure. And I think for me, given the formidable nature, the size of the Chinese forces, they also are better connected internally as opposed to Russia, which is a bit competitive in terms of their various services.

We always thought about China as intellectual property theft and espionage. And now our concerns are much more about Chinese actors pre-positioning for disruptive and destructive activities so that they can delay military deployment, they can get inside the decision cycle for military operations, and to induce societal panic. And what they’ve seen, at least in America in terms of the reaction to Colonial Pipeline, what they see in terms of the reaction to the high-altitude balloon, they probably think it’s not that difficult to induce societal panic that will have Americans say we don’t want to send more money and more manpower and more materiel out to Southeast Asia to deal with this fight.

SK: The size and scale of the Chinese machine is impressive, that they can fire on all cylinders at the same time, and the sophistication of the capabilities has been demonstrated this year. Both of them go back to the principle of working with the private sector. And it’s not just warning the private sector, but it’s also the private sector going hunting on that network and reporting back to us. We cannot go hunting on every private sector network, so we have to share that information for them to go take it seriously, look for what they can find, and tell us what they found so that we can again connect multiple dots.

FP: I wanted to ask about artificial intelligence. How do both of you see AI changing the threat landscape, and how are your respective agencies using AI internally?

SK: We’ve seen cybercriminals use those language models to craft very fancy, very sophisticated phishing emails. They get very personalized very easily when you put them into these large language models. It’s difficult to tie a specific incident to the nexus of a cybercriminal using that language model, but we know of them experimenting with language models and crafting emails, and we need to continue to educate folks on how to spot a fake email.

We’re not strangers to the use of AI—30 years ago, I was using techniques to categorize voice and categorize images using training data. I think it’s how quickly it came on the scene and how quickly it has caught people’s attention. We are busy educating our colleagues in government about the risks of AI, about the precautionary measures that they need to take about putting data into these AI models without proper due diligence, and to recognize that not all AI models out there are created equally. We don’t have full transparency on what goes in some of these models, what training data went into them. We’re not putting any guardrails to block people from using them, but we are putting out some publications to educate government and nongovernment users. It comes with huge opportunities, but at the same time, it comes with some risks.

JE: AI has been around for a long time. It’s really the explosion of large language models, generative AI, that has been progressing hugely. I think you can stipulate there are amazing things that these capabilities will do, but they will also do amazing things for very bad people to enable cyberattacks and all manner of attacks frankly.

We are also very focused across the U.S. government on getting our arms around how to take advantage of the opportunities, ensure we can use the capabilities, and reduce risk to the American people. So the administration is going to put out an executive order here in the next couple months. We are focusing at CISA on the nexus of AI, cybersecurity, and critical infrastructure. So how can we responsibly use AI from a cyberdefense perspective? How can we help assure AI systems, through some of our red teaming capabilities and our coordinated vulnerability disclosure process, and then finally, how do we think about adversarial AI threats to critical infrastructure?

Being the cyberdefender, you have to defend the full enterprise. Bad guys just need to find a way to get themselves in. And so, even if these capabilities allow for stronger cyberdefense, I think the overmatch will go to the offense versus the defense.