Ben Buchanan on AI and Cyber

Summary

ChinaTalk: Ben Buchanan on AI and Cyber

Host: Jordan Schneider
Guest: Ben Buchanan (Dmitri Alperovich Professor at SAIS, former White House Special Advisor to AI)
Release Date: January 11, 2026

Overview

In this episode, Jordan Schneider welcomes Ben Buchanan to discuss the arc of AI and cyber policy, US-China competition in advanced computing, and lessons from the front lines of US government tech policy. Buchanan reflects on his experience in the Biden White House, tracing the evolution of policy around AI, export controls, and the wider technological race, alongside insights into the intersection of AI and offensive/defensive cyber operations. The conversation covers the policy process, the relationship between government and the private sector, and major open questions for the future.

---

Key Discussion Points & Insights

1. The AI Policy Arc: From Hypothesis to Mainstream

• 2021-2025 Retrospective: When Buchanan entered the White House, AI's relevance was clear but unproven; by 2025, computing power’s national security importance is "an established fact." ([00:53])

  “A lot of those things have come true, particularly about the importance of AI to national security and the importance of computing power to AI ... now ... it feels like that has happened.”
  — Ben Buchanan, [00:53]

• Scaling Laws as a Turning Point:
  The 2018-2020 period was pivotal, as the scaling laws (machine capability increasing with more computing power) became central. Buchanan’s 2020 Foreign Affairs piece advocated shifting US focus from data to compute.

  "The real turning point was probably somewhere in the 2018-2020 period when the scaling laws started to come into focus ... that's what's driven a great deal of AI progress." — Ben Buchanan, [02:14]

• Early Policy Wins:
  The Biden administration moved to implement critical export controls and policies before AI hit true mainstream consciousness.

  "One of the things I'm proud of frankly, is that we got some of the biggest action done prior to the whole world waking up."
  — Ben Buchanan, [04:25]

2. Chips Act vs. Export Controls: Strategic Approaches

• Distinction:
  The CHIPS Act was about restoring domestic capacity; export controls were a targeted, security-driven move, not reliant on AGI assumptions. The controls were analogous to previous measures on nuclear or cryptologic technology.

  "I would differentiate between the CHIPS Act and then the export control...The CHIPS Act ... is a supply chain thing ... export controls ... increased the value of chips if they can be used to train more powerful AI systems."
  — Ben Buchanan, [06:35]

3. Balancing National Security and Policy Process

• The “Sullivan Doctrine”:
  National Security Advisor Jake Sullivan’s “as large of a lead as possible” mantra shaped the aggressive-but-not-maximalist US strategy. Buchanan emphasizes trade-offs and respect for decision processes.

  "My view was always a maximalist one, that we should be very, very aggressive. But ... there’s a lot of constraints ... and someone sitting in Jake’s chair has to balance a lot of concerns that a dork like me doesn’t have to balance." — Ben Buchanan, [08:59]

• On Retrospective Policy Choices:
  Buchanan would have acted faster and more aggressively on certain levers (e.g., high-bandwidth memory, earlier parameters for restricting chip manufacturing equipment) if given "do-over" power.

  “Anytime you're doing something that is this technical, I would love to get mulligans and get technical parameters right.”
  — Ben Buchanan, [11:25]

4. Building State Capacity and Enforcement Challenges

• Implementation Concerns:
  Enforcement remains the best argument against ambitious export controls, raising big questions on US state capacity.

  "The best counterargument ... was just, the United States government's not capable of doing this ... the enforcement's not there ... that's the most compelling counterargument." — Ben Buchanan, [15:37]

• Government Hiring:
  Major efforts were expended to "bulk up" with over 1,000 new hires across agencies in 2023-2024 to address this deficit.

5. Government vs. Private Sector: The AI Revolution

• Unique Moment:
  Unlike nukes or space (government-led), modern AI is private-sector driven. The state’s challenge is integrating this outside "revolutionary technology" into its own operations.

  “What is the relationship between the public sector and the private sector at a time when you have a revolutionary technology, probably the first one since the railroad, that is almost exclusively coming from the private sector?" — Ben Buchanan, [19:36]

• Rickover Analogy:
  Buchanan invokes Rickover-style leadership for integrating AI into defense/intel—adapting organizational workflows at scale.

6. National Security Adoption: Three Pillars of AI Competition

• AI Geopolitical Competition:

  • (1) Racing to frontier models (private sector, compute-heavy)
  • (2) Diffusing capabilities globally (winning hearts/minds, markets)
  • (3) Domestic national security adoption (applying AI internally)

    "If we don't use that time, we get zero points. ... I view the AI competition [...] as coming down to three parts. One is the competition to make the best models ... the second is the competition to diffuse ... The third is national security adoption." — Ben Buchanan, [27:36]

• Risks:
  US could build the best models/deploy the most compute but still fall behind if others better integrate AI into their strategic operations, as with tanks in WWII.

  "It is entirely possible that we win the race to the frontier ... but if we don’t get our act together on the national security side, we still fall behind, just as the French and the British fell behind in the early days of the tank.” — Ben Buchanan, [27:36]

7. AI and Cyber Operations: Offense vs. Defense

• Vulnerability Discovery:
  Major recent advances are coming in automated vulnerability discovery/patching, with Google and DARPA pushing the envelope.

  "We are at long last starting to see machine learning systems that can contribute to that work … starting to see evidence in 2025 of that kind of capability." — Ben Buchanan, [33:18]

• Automation Trend:
  The trajectory is clear: more cyber operations will be automated, and “AI paired with the top humans” will likely dominate, but pure AI-driven operations couldn’t be dismissed in the future.

  "There’s a direction of travel that’s pretty clear here, which is towards increasing automation, increasing capability for vulnerability discovery by machines." — Ben Buchanan, [35:49]

• Defense Implications:
  If breakthroughs allow all code vulnerabilities to be instantly detected before deployment, a "defense dominant" world is possible—but this is an organizational challenge as much as a technical one.

8. The Messiness of Policy and Impact of Individual Passion

• Process Lessons:
  Theory is often easier than practice—complex change requires navigating (sometimes frustrating) interagency/government processes.

  “...theory was worked out long before then, but it still was a cumbersome process to get the system to do it. And again, sometimes for good reason.” — Ben Buchanan, [14:34]

• Technical Understanding as Policy Leverage:
  Policy often advances when technical experts personally push ideas, counterbalance lobbying dollars, and make cases at the right time. However, big company lobbying (Nvidia, et al.) is powerful and ever-present.

---

Notable Quotes & Memorable Moments

• On Export Controls' Impact:

  "We’re talking about a company that’s worth hundreds of billions of dollars—Nvidia. We’re talking about very important technology. ... Those are not things that should be done lightly." — Ben Buchanan, [15:05]

• On Enforcement as the Key Challenge:

  "The best counterargument that I never heard to our policies was just, the United States government's not capable of doing this ..." — Ben Buchanan, [15:37]

• On AI's Disruptive Potential:

  "Everything is interconnected. ... That has been the challenge ... teaching it in the classroom and ... in writing." — Ben Buchanan, [61:18]

• On the Role of Individual Technical Passion in Policymaking:

  "I think there's a power law distribution for this kind of stuff ... chip manufacturing ... and AI: that nexus was just by far the highest leverage thing." — Ben Buchanan, [13:02]

• On the Policy Process:

  “Every policy making process is improvisation ... I certainly didn’t feel like I had a sheet of music I was reading from ... I think it has to be jazz.” — Ben Buchanan, [77:34]

---

Timestamps for Important Segments

| Timestamp | Segment / Topic | |------------|-----------------------------------------------------------------| | 00:53 | Retrospective on AI's rise in US policy | | 02:14 | Key breakthrough: Scaling laws and the focus on computing power | | 04:25 | Reflections on waking up the world to AI’s significance | | 06:35 | CHIPS Act vs. export controls—different rationales | | 08:59 | The “Sullivan Doctrine” and policy aggressiveness | | 14:09 | Policy process vs. theory; the reality of government work | | 15:37 | State capacity & enforcement as policy bottlenecks | | 19:36 | The public/private divide in revolutionary tech | | 27:36 | Three parts of AI competition, policy lessons from WWII tanks | | 33:18 | AI's evolving impact on cyber offensive/defensive operations | | 61:18 | On the difficulty of writing/teaching about hyper-connected AI | | 68:00 | Predicting political and societal AI issues by 2028 |

---

Open Questions & Areas for Further Research

• Recursive Self-Improvement:
  Recent results (e.g., Alpha Evolve) demonstrate AI systems accelerating their own improvement.
• Energy as a New Strategic Fulcrum:
  US/China competition in data center energy supply may be the next critical battleground, and the US needs to step up.
• US vs. Chinese Tech-Military Synergy:
  Buchanan wants deep dives into the real nature of Chinese tech companies’ relationship with the state/military.
• Future of Regulatory Framework:
  How can government adapt to regulate AI risks, especially as automation and economic/social transformations accelerate?

---

Book Recommendations

• A Brief History of Intelligence by Max Bennett
  “If you want to take a step back and think about what is it we’re actually talking about when we talk about intelligence ... that was a great book.” — Ben Buchanan, [48:08]

---

Tone & Style

Buchanan brings a combination of technocratic optimism, healthy skepticism, and hard-earned realism from policy trenches. He is both reflective (“I remember what was called China Econ Talk ... we’ve already brought in the aperture here.” [18:48]) and sometimes wryly self-aware (“That’s great, Professor Buchanan, you’ve worked out the theory, but actually what we're doing here is practice.” [14:35]). The conversation is lively, ranging from geeky tech specifics to grand strategic analogies.

---

Closing Thoughts

Ben Buchanan's journey—from early conviction in AI’s strategic role, to hard-fought government policies, to worries about future state capacity—captures the urgency and complexity of current US technology strategy. The episode offers a nuanced look at how ideas become policy, why timing and technical intuition matter, and the persistent challenge of evolving organizations to meet a transformational moment.

---

End of Summary

Transcript

[Jordan Schneider]

Okay. Ben Buchanan, the Dmitri Alperovich professor at sais who served in the Biden White House on many guises, including a special advisor to AI Also, also the Thrice author and Oxford like quarterback. I don't know what his passer rating was, but we'll give him a pass on that. We're going to talk about AI and policy. Ben, welcome to Chinatown.

[Ben Buchanan]

Thanks for having me. Long time listener, first time caller.

[Jordan Schneider]

So. All right, we're sitting here November 2025. It's been a long road. What moments, trends, events stand out to you? Looking back about AI and policymaking since you joined the Biden administration until today.

[Ben Buchanan]

Probably the biggest thing is that there were a lot of things that were hypotheses when we showed up in the White House in 2021, hypotheses that I believed to be clear, but I couldn't prove to anyone. And I think a lot of those things have come true, particularly about the importance of AI to national security and the importance of computing power to AI. And I think you could have drawn good inferences about those things in 2021. You could have said, oh, AI is going to affect cyber operations. It's going to affect US China, competition. It's going to get good enough that it'll really reshape a lot of that stuff. And it'll get good enough because computing power is going to just continue to drive this bus forward as we scale these systems up. But I don't think that was proven in any kind of real Way in 2021. But now sitting here in 2025, it feels like that has happened. And maybe most important, that is going to continue to happen in the years ahead.

[Jordan Schneider]

Maybe there's a lesson there to go back to the sort of 2015-2020 arc. A lot of people think a lot of things are going to be the next thing. Is this just happenstance? Was there some, like, epistemic lesson about the way folks who saw AI as the next big thing sort of identified it?

[Ben Buchanan]

Look, I would love to say when I started getting into AI in 2014 and 2015, I knew exactly where all this was going to go. I think the truth of the matter is I just found it interesting, and it was the kind of thing that it raised really interesting questions about what technology could do. At the time, I was doing a lot of cyber operations stuff, which is interesting in its own right. But fundamentally, cyber operations is about this cat and mouse game between offense and defense. It's cops and robbers on the Internet. And that's. That's good as far as it goes. And there's a lot of really cool stuff that unfolds in that game. But at the time, in 2015 or whatnot, I thought, wow, AI is like at least conceptually driving towards something bigger, asking us questions about intelligence and humanity and even broader than cyber operations in its impact. And that's how I got into it. And then once I started getting into it, it became pretty clear that, wow, this technology is getting better at an accelerating rate and we can, we can kind of draw lines forward on the graphs to figure out where is this going to go. And I think the real turning point was probably somewhere in the 2018-2020 period when the scaling laws started to come into focus. And by that I mean the way in which machine capabilities scaled with additional computing power. And that's when I started to have a conviction that this is going to really matter for international affairs and that computing power was the fulcrum. I wrote this piece in Foreign affairs and the summer of 2020 called the US has AI competition all wrong, which basically says, stop focusing on data, start focusing on computing power. And for the last five years or so, I think the scaling laws have held and that's what's driven a great deal of AI progress.

[Jordan Schneider]

Reflections on sort of different pieces of the broader ecosystem. And like, what, you know, because now the light is on with everyone. Like, this is a front page story all the time. We're living in a giant, you know, Nvidia is worth $5 trillion. Like, like the world has woken up. But kind of looking back, you know, different lights turned on at different times. Is there any, anything interesting about the process by which that happened?

[Ben Buchanan]

I think probably the, the strongest technical light came in 2020 with the scaling law paper from Jariel Amade and the crew that later went on to found Anthropic, which really put some math behind the intuition, I think a few people had held about the importance of computing power in rapidly accelerating AI performance. And then GPT3, which came out in May of 2020, a crazy time in America. Covid George Floyd protests and all of that. GPT3, I think, really put even more evidence to it to say, like, wow, you could make big investments on this and see returns on those investments in terms of machine capability. That was enough for me to have the conviction that I had and I think for other people who went to the Biden administration to have conviction about the importance of computing power. We then spent 2021 and 2022 actually getting the export controls and the like. Into place. Probably the next big thing was after we did the export controls, ChatGPT happened and that, that was I think, you know, October, November, probably November of 2022 when it happened. And that was the thing that I think really kicked AI into the mainstream as you know. And then since then it's just been a parade of even bigger things. The, the Kevin Roose article in the New York Times in 2023 brought it to a new set of, you know, non techie people. And then of course the increasing capabilities since then have, have only accelerated. But one of the things I'm proud of frankly is that we got some of the biggest action done prior to the whole world waking up. And when the, the, when that wake up did happen that we could say truthfully, like we've already done some of the most important policies here, there's much more to do, but we've already taken big steps to being well on our way to, you know, addressing this.

[Jordan Schneider]

It seems like the CHIPS act though was not necessarily a like AGI pills policy from its get go. I'm curious how you feel like that ARC aligns with this stuff.

[Ben Buchanan]

I would differentiate between the CHIPS act and then the export control. So the CHIPS act is the legislative step. I get no credit for that. Drew and Chabra, Safecom, they get, they get a ton of credit for having worked on that. I don't think that's an AGI pill thing at all. I think that is a supply chain saying chips are important for a lot of reasons and we need to have domestic chip manufacturing as we did decades ago, but no longer currently have. So a really good policy. But I don't think it particularly. You know, there's a lot of reasons you could get to that policy outcome without believing in AGI or even really powerful AI systems. Then there's a chip controls, which is what I was referring to. And they're both key fulcrums of Biden administration policy. On the chip control side, I don't think those policies need to have a kind of assumption about AGI in order to be really smart policies. And if you look at how we justified those policies, we talked about nuclear weapons, we talked about cryptologic modeling and all the kind of things you could do with those chips before you even get to really powerful AI systems. Everything in that justification is completely true. I think it's just a robustly good action given the importance of computing power. That said, if you do believe that AI systems are going to get increasingly powerful, which we all did, or many of us Did I guess I should say then the chip controls make even more sense because it just increases what the value of the chips is if they can be used to train more powerful AI systems. So there's a lot of reasons why I think it was a long overdue policy and the right one independent of AGI.

[Jordan Schneider]

So we had Jake Sullivan on recently talking about know the Sullivan doctrine as far as, you know, as big of a lead as possible. But in the way it ended up manifesting was not the maximalist version of as big as possible, you know, as far a lead as possible. When it came to doing controls, there were sort of other considerations which ended up mediating where they ended up falling in October 2022 and when where how they ended up evolving over the next few years. Thoughts? Reflections on maybe first just like the process of making of. Of, you know, bringing these to the. Bringing these to the table.

[Ben Buchanan]

Process started in the. In 2021 when a small group of us got to the White House. I think probably most of us have been on the Chinatalk podcast before. Folks like Tarun Chabra, Chris McGuire Safe Khan, Teddy Collins came later, myself. And we had these convictions, I think, about the importance of computing power. And then Jake honestly gave us a lot of rope, I think, and Jake deserves a lot of credit. And at a time when not many people were caring about AI, when the world's focus was on things like Covid and Afghanistan and in 2022, the war in Ukraine, Jake and the senior White House staff heard us out on this and eventually in 2022 we got to the point where we were actually going to do it. So, you know, everything in the government's a slog sometimes and this was an interagency process. But I think something like this shouldn't be done lightly. And I think it's good that there's at least some process to adjudicate debates and so forth. As you mentioned, Jake gave a speech in September of 2022 where he talked about as large of a lead as possible in some areas. And probably my view was always a maximalist one, that we should be very, very aggressive. But I also recognize that there's a lot of constraints on possibilities and someone sitting in Jake's chair has to balance a lot of different concerns that a dork like me doesn't have to balance. I'm just focused on AI and chips and this kind of techy stuff. So I think everyone can draw their own conclusions about what we should have done when and so forth. But I think I am very proud that we got the system to do this even before AI was the mainstream thing that it quickly became.

[Jordan Schneider]

Yeah, I mean the one hypothetical that he entertained was the do FTPR on semiconductor manufacturing equipment just from the get go. So you don't end up having this kind of like, okay, we list this company and then they have a subsidiary and then S1 Fab but the other Fab across the street isn't listed. And ultimately you have this really dramatic chart of semi cap actually doubling the amount which is ends up being exported after the controls come in place. Is that the big fork in the road? Like what are, what are the other. What else is kind of contingent when we're looking at how China can manufacture chips today that we don't.

[Ben Buchanan]

It's probably fair to say that on chip manufacturing equipment that would have been the more aggressive option is to, rather than cutting a deal with the Dutch and the Japanese to use the foreign direct product rule to, to basically blanket ban the chip manufacturing equipment. Going to the country the way we did with chips. That's probably one option. I think, you know, if we were doing it all again, probably would have been more aggressive earlier on things like high bandwidth memory and the like. Or we would have had a different parameter, the primer we used in 2023 related to performance density of the chips we would have had in 2022. So anytime you're doing something that is this technical, I think I would love to get mulligans and get technical parameters right. But I think the core intuition and motivation for the policy has held up pretty well and most of the execution has been pretty good, at least from a policy perspective. So I'm not, I wouldn't second guess a lot of it. I wouldn't change a lot of it except to say, yeah, sure, I would have loved to have done even more stuff even faster. But that was my disposition all along in this process.

[Jordan Schneider]

I want like more macro lessons. So it's just trust the nerds who are really excited about their first few books, like, you know, there must be. Is there anything repeatable about the fact that you did have a team that was focused on this? You know, back when Nvidia was worth a lowly 500 million, you know. You know, $500 billion.

[Ben Buchanan]

I think it was something I remember thinking about in the White House. And Jason Matheny in the White House was a person who asked this question very well, which was sort of like, okay, we found this one. How many other things like this are there out there? Can we do this for 10 other things. And we did do something similar eventually in biology and biology equipment and the like. So I think there probably were others. But I think it's also the case that there's a power law distribution for this kind of stuff and semiconductors and chip manufacturing equipment and AI. That Nexus was just by far the highest leverage thing. And I'm glad we found it. I'm glad we did it when we did it. But I think, at least I don't know of another thing like this at that level of scale. There probably were others that were less that we could have done and some that we did do. But this was, this was, I think, the biggest, highest leverage move available to us.

[Jordan Schneider]

Other, like statecrafty stuff. I don't know. What'd you learn about the way the world works sitting as a special advisor to AI in the final few years?

[Ben Buchanan]

I learned a lot about process. I think I had a concept that, you know, someone, maybe the president just makes a decision and then it all happens from there. And I think anyone who's worked in the government can tell you there's a lot more process involved. And some of that process is good and some of that process is annoying, but I think there's just a mechanism to it that is important. And I recall a moment when.

[Jordan Schneider]

I.

[Ben Buchanan]

Made some point in a meeting and someone else said, well, that's great, Professor Buchanan, you've worked out the theory, but actually what we're doing here is practice. And it's like, yeah, you're right, that is what we're doing here. And it turns out in many cases the theory is actually not that difficult. And, you know, a lot of us had written about some of the stuff in 2019 and 2020, and the theory was worked out long before then, but it still was a cumbersome process to get the system to do it. And again, sometimes for good reason.

[Jordan Schneider]

Why?

[Ben Buchanan]

Well, we're talking about, you know, I don't know what the export market was at the time, but we're talking about a company that's worth hundreds of billions of dollars in video. We're talking about very important technology. We're talking about essentially cutting an entire, the largest country in the world by population off that technology. Those are not things that should be done lightly. I think it is fair that there should be a gauntlet to be run before the United States takes a decision like that.

[Jordan Schneider]

You know, what are your sort of like Jen Palka state capacity takes after doing this stuff?

[Ben Buchanan]

I think there's real questions on enforcement and the best counterargument That I never heard to our policies was just, the United States government's not capable of doing this, that maybe we could write the policies eventually, but the enforcement's not there. There will be subsidiaries bis the Bureau of Industry and Security and the Department of Commerce that carries out the enforcement's chronically underfunded. And I think, I don't buy that argument. I think the US Government should do this and could do this. But I think that's the most compelling counterargument, more so than anything about the core theory of the policy is wrong. It's just we're not capable of this. And I think I'm all for building out state capacity and basically every aspect of AI policy. And when I got to one of my later roles in the White House of working with the chief of staff's office and the domestic side of it as well, where I had a little more control over that, this was a big priority for us. And we hired, you know, probably more than a thousand people in 2023 and 2024 across a large variety of agencies to try to build that state capacity.

[Jordan Schneider]

Maybe the question is like, if you had the level of maybe not 100% of the level, but like maybe 65% of the level of top cover that like Doge had in the first few hundred days to like, take big swings, not really worried if you're going to get sued two years from now. You know what, Like, I know you're going to say rule law is important, but like, like, if you had your druthers and things would all work out fine. Like, what are, what are, you know, directions you would have liked to have run harder on?

[Ben Buchanan]

Well, I think rule of law is important, Jordan. So that's one, two. I think it's actually easier to burn things down than it is to build them up. And I think we had a lot of top cover. I think Jake Sullivan, Bruce Reed, ultimately the President gave us a lot of top cover at every turn. But on the China competition front, probably at the time, I would have wanted to have done more of the things faster and more aggressively, especially given what I know now and how the general theory was. Right? So, you know, you mentioned chip manufacturing equipment. That was one. Talk about high bandwidth memory, that's another. That didn't come till a couple of years later. Obviously I would have bulked up if I had this kind of dictatorial control, I would have bulked up enforcement capabilities and the like. I think now there, there's a lot that, you know, a lot of that stuff still holds up the, the China committee in the House did a good report maybe a month or two ago on some. Some of the things that could be done on chip manufacturing equipment. Those, I think, are robustly good things to do. Basically, we should be doing them as soon as possible. If we could have done them earlier, that would have been great, but certainly we should be doing them now. And stuff like that is in the Trump AI action plan. This is not a partisan thing. They just haven't done it.

[Jordan Schneider]

Okay, sorry, taking the camera back, just from the. I know it's called China talk, but, like, it's not really Chinatalk anymore. On the Jordan.

[Ben Buchanan]

I remember what it was called. I remember what was called China Econ talk. So I feel like we've already brought in the aperture here.

[Jordan Schneider]

You know, it's like a billion people is not enough for me. We got it. We got to do it just on the broad prompt of, like, harness the opportunities, manage the risks. Like the. The special advisor for Ben Buchanan juiced up with some Doge cocktail of operational freedom. I mean, I don't know, like, what, what are you, like, setting. Setting, like, what Trump is going to do aside, what do you think that the sort of federal government is capable of Just like, like, what do you think? What do you think the federal government could really do if they put their mind to it?

[Ben Buchanan]

So putting putting China aside and wearing my AI hat more than my China hat, I think the most fascinating question of the moment is what is the relationship between the public sector and the private sector at a time when you have a revolutionary technology, probably the first one since the railroad, that is almost exclusively coming from the private sector? And you can think about nukes and space and all this other stuff, it's coming from the government. Maybe the private sector is doing the work, but the government's cutting the check. So I think this is a question that we just started to get our hands around, but that definitely, if I had this level of control you're talking about and I were still in the government, is one I'd spend a ton of time on, and probably I'd be going to places like DOD and the intelligence community and saying, you have to find ways to develop this technology and build it into your workflows and take what the private sector has built and really make sure we are using this for full national security advantage. And, you know, I actually think the analogy there is maybe less like Doge, though there's some of that and more like, who's the Rickover of this era? And what does that look Like Rickover taking stuff that the military was building. But what does the Rickover look like for AI? Someone who's taking the technology and really integrating it into military operations? Or what does it look like, you know, the CORONA program and the American spy agencies did incredibly impressively pushing the boundaries of the technological frontier. The current program being the one where they took basically early spy satellites and dropped the film casters from space. Just insane that it worked. That's the kind of stuff that I think requires a lot of air cover, a lot of money in some cases and a lot of ambition. So I would be really pushing, and we did push to get government agencies to do that kind of work. Similar levels of ambition, taking a private sector developed technology and putting it to use for their very important missions.

[Jordan Schneider]

Yeah, I mean, I guess the question is like, is there too much structural bounds on doing Rickover type stuff for, you know, the situ, like the national security complex as like currently established to take those, those big swings?

[Ben Buchanan]

As someone who's never worked in DoD or the IC, I don't know that I have a high confidence view. But what I did see is that the answer probably is yes. And there's a line, we worked on the President's national screen memorandum on AI, and there's a line in the introduction of that document which says something like this is not just about a paradigm shift to AI, but this is about a paradigm shift within AI. And I think if you go to DOD or you go to the intelligence community, a lot of folks will say no, no, of course we do AI. We've done AI for a long time. Don't you know we funded a lot of AI research in the 1980s. But really what we're talking about is saying how quickly after, you know, Google drops Gemini 3 or anthropic drops Claude 4.5, how quickly can we get that into intelligence community and duty workflows, including classified spaces, and put it to use for the mission? How much can we redesign those workflows to accommodate what the technology can do? In the same way that in the early days of the Industrial revolution, everyone had to redesign factories to account for the engines and electricities and like of the day. So that's the kind of stuff that I think is important. To be clear, I'm not saying I'm qualified to do any of that, but I think that's the kind of if you're really driving top down change, that's where I put a lot of focus if I want to benefit American national security.

[Jordan Schneider]

Yeah, because this is the the, the thing is like, like private sector firms will be able to outcompete other private sector firms by doing a better job of employing AI and whatever capabilities it unlocks. So if that is automating low level stuff, if that is informing sort of like strategic C suite decisions, if that's whatever in between, like you have a sort of natural creative destruction element going on. I mean, as Sam Altman said at one point, you know, if, if OpenAI isn't the first, doesn't, doesn't get. Kick its CEO out of its job, isn't the first company in the world to kick its CEO out of a job and hand the reins over to AI, then like we're doing something wrong. And it is going, it is inevitable that governments all around the world are going to be slower adopting that than like, you know, the five person startup that's worth $5 billion because they can, you know, be incredibly nimble and are really technically proficient in working at and even beyond the frontier of like what is commercially acquirable. But I guess the sort of, the question is, aside from like people sitting in the White House telling agencies to get their shit together, like what are the. Or like just being scared of being out competed by China or you know, Mexican cartels or whatever, like what is the, what could the forcing function be to drive some of the sort of legislative and executive branch action to have that stuff actually happen?

[Ben Buchanan]

Yeah, I think there's three points here. The first is the stakes are higher for DoD in the intelligence community than it is for the five person startup. So it is reasonable that to some approximation those places would go a little bit slower because we're dealing with life and death and not cat yoga or whatever the startup is these days. Second, the forcing function to students of history should be what you said, which is the fear of being out competed. And you know, Jordan, you have sent me enough books on World War II over the years to know that the tank offers a very kind of illustrative analogy here and that it was the British and the French who invented the tank at the waning years of World War I. They didn't really know what to do with it, they didn't know how to apply it. And then it was the Germans in the early days of World War II who figured out how to, how to use it. And that is the kind of thing that I think offers a lesson of like, well, it is blitzkrieg and it is, you know, this technology that was invented at the end of World War I that kind of sits dormant that the Germans pick up and then they use it to just roll across Europe. I am deathly afraid of that happening in AI where it is America that invents this technology, the American private sector, but it is other nations that figure out how to use it for national security purposes and create strategic surprise for the United States. So that should be the forcing function. I think, realistically, you are going to need significant DoD leadership and intelligence community leadership to drive that. I'm worried we're going in the wrong direction. Laura Loomer got Vin Nguyen, the chief AI officer at nsa, who's one of the best civil servants I ever worked with, she got him fired, so forth. So I'm worried we're going in the wrong direction on that front. But I do think that is the.

[Jordan Schneider]

Imperative and the sort of corollary of that which makes it more scary. I'll give you my take. Curious for yours. Is that the lead that America has in compute, there is a world in which that means that we can get away with not doing a good job on the sort of, like, operational level reimagining of intelligence and defense. But there's also many futures in which, okay, you know, America ends up having 2x compute or 3x compute. But the sort of, the downstream creativity when it comes to using, employing that compute for national security purposes is such that, like, you can't just rest on your laurels of having more data centers. Um, so, you know, we're. We're not just cool. We're not like, just good. Because Nvidia makes better, better and more chips than Huawei.

[Ben Buchanan]

Emphatically not. And I think, again, I'm very proud of our policy. But even the best defense of our policy to buy a lead or build a lead over China in terms of computing power is to say it buys us time. And then if we don't use that time, we get zero points. It's not like, oh, well, you get a B plus because you built the lead and then you blew it. You still blew the lead. So, yes, I think. I think basically I view the AI competition with China as coming down to three parts. One is the competition to make the best models. The frontier. This is where compute really helps. The private sector is taking the lead. The second is the competition to diffuse those capabilities out into the world to win the global market, to win over developing nations and the like. We haven't talked about that. We can get there. And then the third is national security adoption. To say, okay, we're going to take this technology that we're inventing that only we are inventing at the frontier, and we're going to put it to use in our national security missions. It is entirely possible that we win the lead to the front, we win the race to the frontier, we have success in that competition. But if we don't get our act together on the national security side, we still fall behind, just as the French and the British fell behind in the early days of the tank. So, yes, basically, I agree.

[Jordan Schneider]

And I think the. The other thing folks don't necessarily appreciate is that if you just win a I or you win part A and part B, it doesn't solve everything. Like, there are always other moves you can do if you feel like your adversary is winning on this dimension of the conflict. You know, like, data. Like, okay, America has 10 times more data centers. What happens when the lights go out? Or what happens when some drones fly into them? I mean, there's just, like, so much asymmetrical response. You can have that to bank your entire future on. You know, super intelligence seems like a rather foolhardy strategic construct.

[Ben Buchanan]

Yeah. I think two things are true here, which is, I would never advise a nation bank its entire future on superintelligence. On the other hand, I would never advise a nation cede preeminence in AI. Like, preeminence in AI is a very important goal for a nation and for the United States in particular, and shows up in all parts of economic and security competition. But definitely it's not the case that, like, oh, we have more data centers and we somehow we've cut China off from chips. We're good. That is the beginning of the. The competition. It is far from its end.

[Jordan Schneider]

All right, so let's do a little case study. Your first two books, the Cybersecurity Dilemma. Bestseller, the Hacker and the State, which we were almost going to record a show on until Ben, you know, got a job. We're all about cyber. So what's the right way to conceptualize the different futures of how AI could change the dynamics that we currently see?

[Ben Buchanan]

I think. I think the intersection of AI and cyber operations is one of the most important and one of the most fascinating something I've been writing about for a long time. And there's a bunch of different ways you could break it down. Probably the simplest conceptual one is to say we know what sometimes called the kill chain, that basically the attack cycle of cyber operations looks like. We know what the defensive cycle looks like just for each of those steps. How can AI change the game? And there's been so much Hype here over the years and we should just acknowledge that at the outset. But also there is a reality to it and as these systems continue to get better we should expect the game of cyber operations will, will, will continue to change. You could break that further into two parts. If you look at the, the offensive kill chain, I think you could say one key piece of this is vulnerability, discovery and exploitation. And that is a key enabler to many those throwing out all cyber operations. We've seen some data that AI companies like Google are starting to have success doing AI enabled program analysis and vulnerability research and the like. In the way that was just not the case a few years ago. The second one is actually carrying out offensive cyber operations with AI moving through the attack cycle more quickly, more effectively with AI. We can come back to that, but let's stick with the vulnerability for a second. When I was a PhD student a postdoc DARPA ran something called the Cyber grand challenge Las Vegas 2016. Early attempt to say could machines play capture the flag at the DEFCON competition, the pinnacle of hacking. And the answer was eh, kinda. They could play it against each other but they were not nearly as good as the best humans. This was so long ago. We weren't even in the machine learning paradigm of AI. And then when I was in the government and we were looking for things in 2023 to do on AI, I was a big advocate of creating something called the AI Cyber Challenge, which essentially was the Cyber Grand Challenge again and saying now we're in a different era, machine learning systems and the like, what could be done. And DARPA ran that in 24:25 and I think that that told us a lot about, there probably is something there about machine learning enabled vulnerability, discovery and either patching or exploitation. So that's probably where I'd start.

[Jordan Schneider]

Okay, yeah, let's, let's, let's, let's follow your framework. Let's start on the offensive side of the divide that you gave. What is the right way to conceptualize what, what consists of offensive cyberpower and how does AI relate to those different buckets?

[Ben Buchanan]

At its core, offensive cyber power is about getting into computer systems to which someone does not have legitimate access and either spying on or attacking those systems. So a key part of that is this vulnerability research that we were talking about finding an exploit in Apple iOS to get onto iPhones or in critical infrastructure to get onto critical infrastructure networks. And I do think this is the case where we are at long last starting to see machine learning systems that can contribute to that Work. I don't want to overhype this. We have a long way to go. But it is the case, I think that Google has used, their thing is called Big Sleep. And Google has used AI systems to find significant zero day vulnerabilities. Now they're using the systems to patch those vulnerabilities as well. But I think we're starting to see evidence in 2025 of, of that kind of capability. And it's reasonable to expect, I think that this is the kind of thing that nations will, if they're not already interested in, will before long be interested in because of how important that vulnerability discovery capability is to offensive cyber operations. So that is a key part of national power, I think, insofar as cyber is a key part of national power, getting access to AI systems that can discover vulnerabilities in your adversary networks.

[Jordan Schneider]

And presumably this just comes down to talent. Just like how many good folks can your government hire and put on the problem?

[Ben Buchanan]

I think, I think before you get to AI, it definitely comes down to talent. And these are some of the most important people that work at intelligence agencies for those who can find vulnerabilities. And it's a very, very cognitively demanding, intricate art, the argument goes. Again, I don't want to overhype it, but the argument goes, well, I can start to automate some of that and I think to some degree that will be true. And to some degree you'll still need really high end talent to manage that automation and to make sure it all actually works.

[Jordan Schneider]

Yeah, I mean, I guess the question is like, so it's talent and it's money, right? Because you can buy them as well. So it's, you know, I guess we're left with a TBD like we are in many other professions, thinking about to what extent, you know, the AI paired with the top humans is going to be more powerful, whether it allows you to know more entry level people, to be more expert, or whether we'll just be in a world where the AI is doing the vast majority of the work that was previously a very, you know, artisan endeavor.

[Ben Buchanan]

Yeah, I mean, I think, I think this is, I think, I think this is tbd, but I also think there's a direction of travel that's pretty clear here, which is towards increasing automation, increasing capability for vulnerability, discovery by machines. And we should expect that to continue. I don't see any reason why that wouldn't continue. We can debate the timelines and the pace and so forth, but I don't see any reason why it Wouldn't continue, It is worth saying that might not be a bad thing. And in a world in which we had some hypothetical future machine that could immediately spot insecure code and point out all the vulnerabilities, that would be a great thing to bake into Visual Studio and all the, you know, development environments that everyone uses. And then we'll never ship, the theory goes, will never ship insecure code again. So it is totally possible that this technology, once we get through some kind of transition period, really benefits the defensive side of cyber operations. Rather than the.

[Jordan Schneider]

Okay, staying on the offensive side, though, let's go to the exploit part. Okay, I'm in. I've been Ben's phone. I don't want to get caught. I want to hang out there for a while and see all the doordash orders he's making whatnot. Is that more or less of an AI versus a human game?

[Ben Buchanan]

I think so. Just to make sure we're teeing the scenario up here, what you're talking about here is you have a vulnerability in a target, you've exploited that vulnerability, you're on the system, then you want to actually carry out the operation. And can we do that autonomously? And this is where we are starting to see some evidence that folks are already doing this, that hackers are already carrying out offensive cyber operations in a more autonomous way. Anthropic put out a paper not too long ago where they attribute to China a set of activities that they say autonomously carried out key parts of the cyber operation. It's worth saying here, as a matter of full disclosure, I do some advising for Anthropic and other cyber and AI companies. I had nothing to do with this paper, so I claim no inside knowledge of it, but I think it's fair to say OpenAI has published threat intelligence reporting from earlier this year as well, about foreign hackers using their systems to enable their cyber operations. So there is starting to be some evidence essentially, that AI can increase the speed and scale of actually carrying out cyber operations. That truly makes sense to me.

[Jordan Schneider]

So I guess the question is, like, on the, you know, we have a rough parallel of the. Of the offense side and the defense side is, for the offense side, you want to find the vulnerabilities. On the defense side, you want to patch the vulnerabilities. And on the offense side, once you find your vulnerabilities, you want to, you know, exploit them and do all the snooping and hacking or whatever. And then on the defense side, okay, once someone's in your system, you Want to be able to find them as fast as possible and kick them out. I'm curious, is there any, is there any reason to think that there would be some like, different coefficient attached to how well I can find the vulnerabilities versus, like, make use of them, or should we expect this sort of development to happen roughly in parallel?

[Ben Buchanan]

I think it'll roughly be in parallel. I do think there's a world in which we get to, if we play our cards right, we can get to a defense dominant world. Because if we had this magic vulnerability finder, we would just run it before we ship the code and that would make the offense's job much, much harder. And Chris Rolfe of Meta and a bunch of other places has done good writing on this subject, I think, and has made the case for it most forcefully. But we have to get there. And there's so many things in cybersecurity that the best practice would solve the problem, but no one does the best practice or not enough people do the best practice. And that's why cybersecurity continues to be an industry, is because it's this cat and mouse game, as I said at the beginning. So I am, I am cautiously optimistic that we can get to a better world because of AI and cyber operations, offensive and defensive. But I'm very cognizant we're going to have a substantial transition period before we get there.

[Jordan Schneider]

I guess maybe my question is, like, is like, are there countries today that are really good at one half of the equation but not the other?

[Ben Buchanan]

Well, I think America, you know, obviously there's limits of what we can say in the setting about offensive cyber, but I think America has. Has integrated cyber well into singles intelligence and the like. And.

[Jordan Schneider]

Oh, I'm sorry, I meant on, on the split between finding, finding the. Finding the exploits and using the exploits. Like, is that basically the same skill or the same, like.

[Ben Buchanan]

I think they're very highly correlated. Yeah, I think, if anything, if anything, I think using the exploits is easier than finding them. And finding them is a very significant challenge. And there's not that many found per year and all that kind of stuff. But there's a notion we have in cyber security of the script kitty, which is someone who can take a essentially off the shelf thing and just use that themselves without really understanding how it was made. So, yeah, I think that that's the difference.

[Jordan Schneider]

And then, yeah, I don't know. Yeah, the net assessment on the, on the defense side, I think it's worth.

[Ben Buchanan]

Just saying that on the defensive side, huge Portions of cyber defense are already automated with varying AI technologies. And the reason why is the scale of what we ask network defenders to do is so big that you need to have some kind of machine intelligence doing the. The triaging. Otherwise it's just gonna be impossible. And this is a huge portion of the cybersecurity industry. It's a huge portion of things as basic as spam filters and things that are more complex and intrusion detection and the like. And the, the picture you painted before about you kind of have this race between offense and defense, and both sides are using machine learning in the race. I think that's basically right. I think it's even more fundamental to the defensive operations than it is to the offensive side of the ball.

[Jordan Schneider]

Gotcha. Broadening out theories of change for policy. What inputs matter and which ones don't.

[Ben Buchanan]

In the current Trump administration or just more generally?

[Jordan Schneider]

Yeah, more generally. Well, I guess, like we've already talked about one is like, like individuals who are really passionate about a thing and then get into the government and then convince their principals that their thing is important. But there are other. Clearly are other things going on besides like staffers, passions. Right. That end up.

[Ben Buchanan]

Yeah. And you shouldn't win policy fights based on passion. You should, you know, you should bring some data. I think on subjects like technology policy, there is, in a normal administration, there is still a lot of alpha in actually understanding the technology or if you're in a think tank, teeing up an understanding of the technology for the principle because it is really complicated. And if you're looking at something like the chip manufacturing supply chain, there's just so many component parts and tools, and that's, that's probably the most complicated supply chain on earth or close to it. So I think this is a case where technical knowledge, either on the part of the policymaker or on the part of a think tank author or the like really is just a huge value above replacement. And when folks come to me, my students and others come to me and say, well, what kind of skills should I develop such that I can make contributions to policy down the line, either in the government or advising the government, that is almost always my answer is get closer to the tech.

[Jordan Schneider]

It's kind of a bigger question, though. I mean, like, there's money, there's, you know, news reporting, there's. How can I frame this? Like, like that answer still is kind of on the like, okay, what should you do as an individual? I just kind of reflecting on the, the way debates have gone over the past five years around this like what is the, what is your sense of the pie chart of the different forces that act on these types of questions?

[Ben Buchanan]

Well, certainly other forces include money and include lobbying and, and inputs from corporations that have vested interests. And to some degree that's legitimate and part of the democratic process. And to some degree that can become a corrosion of national security interests. So I think we were able to push it back on that a fair amount and I think our record shows that. But it's undeniable that that is a very key part of how the US government makes its decisions is just the incoming and lobbying from people who have a vested stake in what those decisions turn out to be.

[Jordan Schneider]

Again, the answer you gave is, the one that we want to hear on Chinatalk is like, oh yeah, you just learned the thing and it'll be good. But what, I don't know what else, what else grinded your gears then?

[Ben Buchanan]

I am worried that maybe I'm presenting too rosy of you to China talk friendly of you, but that was kind of my experience. Again, the process was longer than I would like and so forth. But big companies, Nvidia chief amongst them, were not happy about the policies that we put into place. I get that. But also the policy stuck and there's, there's becoming a bipartisan consensus on this that, you know, even lobbying has not been able to overcome. So this is the case where I do think with important exceptions, the facts have, have mostly won out and I think that's good. Now there's probably a lot of aspects of national security policymaking where that's not the case that I didn't work on, but I feel lucky that I'm speaking in my experience here and for the most part my experience has been fair minded people in the government heard us out and made the right decision.

[Jordan Schneider]

What are the other big questions out there? What do you want the kids to write their PhDs on?

[Ben Buchanan]

Well, one of the most important questions at the moment is just how good is AI going to get and when? And that is something that I think, I don't know if that's a chinatalk question, but you said it's broader than just China. So that I think is an important question. And I see no signs of AI progress slowing down. If anything, I think AI progress is accelerating. One of the really interesting papers from earlier this year, something called Alpha Evolve from Google and this provide the best evidence we've seen thus far of what we call recursive self improvement. So AI systems enabling better and faster generation of the next generation of AI systems. So that I think is really significant. And in that case, the AI system discovered a better way of doing matrix multiplication, one of the core mathematical operations in training AI. No one in humanity expected this. We've done matrix multiplications the same way for the last 50 plus years. And this, this system found a way to do it, I think 23% better. So that, that kind of stuff suggests we are at the cusp of continued progress in AI rather than any kind of meaningful plateau. So that, that is one. Another subject that maybe is a little bit closer to the China Talk reader is energy. And you know better than I do the way in which China is just crushing the United States on energy production, which of course is fundamental for AI and data centers. This is an area where I expect the Trump administration to be much better than they actually were. They talked a very big game. Republicans in general are kind of pro building and so forth, but Trump has cut a lot of really important power projects, basically because they're solar projects. Michael Kratzios, Trump's science advisor, said, we're going to run our data centers on coal. That's obviously not realistic. So I do think there's a sense in which that is another fulcrum of competition with really clear application to AI between the United States and China.

[Jordan Schneider]

You got book recommendations? I don't know. What have you, what have you, what have you been reading nowadays?

[Ben Buchanan]

Yeah, I read, I read a book recently called A Brief History of Intelligence by Max Bennett, came out a couple years ago. And I thought that was just a fascinating book in thinking about intelligence, because it's not about AI, it's about basically how human intelligence developed. But you can see over hundreds of millions or billions of years, depending on how you count the development of intelligence, and you can see how evolution was working through a lot of the, the same ideas that humans had to work through when we were developing AI systems over the last 70 or so years, in some cases picking many of the same solutions to some of the same or similar problems. So that is a, that is a book that, if you, if you want to take a step back and think about what is it we're actually talking about when we talk about intelligence. So much focus is on the artificial part. Let's put some focus on the intelligence part. That was a great book.

[Jordan Schneider]

I feel like I would have trusted that book more if it came out in 2020 or 2019. It's just like, well, I don't know the field, right. And there was a whole lot of just so, oh, look how These models actually worked like the organelles, I think.

[Ben Buchanan]

I mean, sure, there's some of that, I guess, but I think the bigger point is just put aside the analogy to AI if you want. It's just a really interesting story of like, how our own brains developed and how human intelligence developed. Now, I don't know enough of the neuroscience to say maybe there's a great rebuttal to it. That's like a different theory. But I found that history of intelligence development in the biological sense really interesting.

[Jordan Schneider]

Okay, I can ask you more questions.

[Ben Buchanan]

But for the ChinaTalk reader and analyst, I think one question that's important is what's the relationship between the Chinese state and the Chinese tech industry? And we talked a little bit earlier about how much of a challenge it is to get the US private sector and public sector work together, at least kind of canonically, it is easier for China to achieve that. I would love to know the degree to which that's true in practice. And to what degree are companies like Alibaba and Tencent and Baidu and Deep Seq working with the PLA or working with the Chinese state, or to what degree are they creating some space for themselves? There was some media reporting a week or so ago. I forget exactly about Alibaba work. I forget which part of the military apparatus it was with. But there's at least some suggestions in media about this. But I would love the chinatalk treatment of the subject.

[Jordan Schneider]

Yeah, I mean, my two cents is like, it'd be weird if they weren't. Right? I mean, like, I think it's fair to say that Microsoft and Google are part of the American military industrial complex in one way or another, at least on the cyber side, to be sure.

[Ben Buchanan]

I mean, we saw particularly on offensive cyber.

[Jordan Schneider]

Well, I think the Ukraine anecdote is a pretty straightforward run about all the work that they ended up doing to try to. Or maybe, I mean, probably more on the. On the defense side. Right.

[Ben Buchanan]

Yeah, I think I would draw a distinction because those companies are in the defensive cybersecurity business.

[Jordan Schneider]

Yeah.

[Ben Buchanan]

But yeah, I mean, I would love to know more about a company like Tencent, which is on the 1260H list, basically identified as working with aiding the Chinese military. I forget the exact designation. So I think chinatalker years will be well served by a deep dive into those kinds of companies and what they're doing for the state over there.

[Jordan Schneider]

Sort of reflecting back, I think it's fair to say that the story of export controls was that it took a lot of sort of political appointee expertise to come in and like be the subject matter experts. We've had a lot of shows and there have been a lot of papers written about how to sort of build in more of a long term like analytical body to serve both, you know, Congress as well as the executive branch to get, get in front of this stuff. So you don't necessarily need CSET to exist, to pay people to do it for you. What are your reflections on the sort of ability for the government to grok emerging technologies? And I don't know, how would you structure this thing?

[Ben Buchanan]

I think it's nascent and I think it got better during the four years I was there. I am worried it is getting worse and I'm worried we've bled a lot of talent from the intelligence community and some of the people who I thought were the sharpest at understanding this technology are no longer there. So that is a concern of mine. The analogy that I often drew upon was if you think about the early days of the Cold War, the United States and Soviet Union were each starting to push into space and spy satellites and all of that. We built entire agencies essentially out of, you know, whole cloth to do that analysis and, and build those capabilities and the like. Getting our, getting our own intelligence capabilities up there and then understanding what the Soviets were doing, that was a totally new thing. And I think we basically have to do something like that here. Now I'm not saying it's a new agency, but I do think it's that magnitude of community wide change to respond to just a completely different technical game than the IC is used to playing or historically has been used to playing. And I think we were lucky to work with a fair number of folks in the IC who at leadership levels got this. David Cohen at the CIA is one example. Avril at ODNI is another. Charles Luftig at ODI is another. So there were people who got it. I think it's just a question of time and consistent leadership. And you know, the President signed a national security memorandum in October 2024 that really provide a lot of top cover and direction. And then we were all out by January. So I don't know, I don't know what the status is now, but I do think it is a big change that's required at the magnitude of what we did during the Cold War to, to extend the reach of intelligence to space.

[Jordan Schneider]

Yeah, I mean it's, it's tricky though because even the space analogy, like that's a discrete technology, like you can kind of like all Right. Like someone's going to have to build the satellites and then we're going to give the photos to the people who know something about Russian missiles and figure it out. But what the sort of technological overhang that AI is presenting is you have this tactical and operational stuff around our conversation with, with cyber, but there's this broader, like, how do you set up an organization? Right. And I feel like the sort of amount, you know, again, if it really hits the amount of sort of job descriptions that are going to change and the ways that private sector companies are going to like evolve in their workflows has the potential to be extremely dramatic. And there is very little in the sort of like regulatory or bureaucratic structure that gives me a lot of confidence that like, that just like having a sort of body over there is going to do it and that these organizations have enough like capacity for renewal, internal renewal, to, to, to, to really do the thing.

[Ben Buchanan]

Yeah, I basically agree. I mean, I think the answer I gave you was the answer about how does intelligence community confront the technology itself that is different from the question of how do they confront their own way of doing business. Yeah. And you are right that AI will and should change key parts of organizational structures, including in the intelligence community, in a way that space fundamentally did not. And I think it is fair to say we articulated that question and sent like the very beginnings of gestures of an answer to that question. But first of all, the tech wasn't there in 23 and 24 when we were really working on a lot of stuff. So there's only so much you can skate to where the puck is going. But it is something that if we were in now, I would hope we were spending a lot of time on.

[Jordan Schneider]

I had this conversation with Jake Sullivan about experience and you know, I asked him something like, on what dimensions did you get better in this job in year four than you were in year one? And on the one hand he was like, I was fucking burned out. I needed a six month break somewhere in there. But also he was like, look, if you're in, you know, living through crises, like being in this, like there's just, there's no way to simulate it. And then I got to thinking, um, you know, we're not that far from a world where I can tell GPT7 to build me a VR simulation of being Ben, Ben Buchanan in the summer of 2021 and try to like, you know, send some emails and talk in some meetings to convince people to do FDPR on semiconductor manufacturing equipment from a sort of Future policymaker education perspective, like beyond doing PhD think tank, reading, writing, analyzing stuff. Like what skills, what, what other skills would you have wanted to have. Have come in? And is there a world in which I can help serve as that, like educational bridge to allow people to operate at a, at a higher octane than they, than they would be going in cold?

[Ben Buchanan]

The first half of the question is very easy. The second half the question is very hard. The first half, the question, you know, essentially, where did I get better over four years or you know, what skills they wish I had that I didn't have in 2021. It's just understanding how the process works, understanding how the US government makes decisions, understanding how you call people, how you run meetings, how you put together an interagency coalition. I was very lucky that I got to learn from some of the just best people on earth in doing that. I think Tarun Chabra is the obvious archetype Chinatalk Guest himself. So that was the skill that I did not have going in. I felt confident on the technology side, but did not have going in. And then I think when I left, I felt much more confident, like, okay, I've learned this. How could you learn that on the front end? I don't know if it's an AI thing. I guess you could, you could maybe do it. But there probably is something in there about, you know, role playing to me always felt kind of hokey, but like, how would you role play this and how do you get people to practice this skill and so forth? I don't know. Maybe there's something there. I hope there is because it'd be great if our policymakers could hit the ground running on that skill in a way that I definitely did not. But I don't know what it looks.

[Jordan Schneider]

Like you've had a year or a little less. You've had coming up on a year now to just have more time playing.

[Ben Buchanan]

It feels longer than a year. Jordan, I can tell you that it hasn't been the fastest year of my.

[Jordan Schneider]

Life to just play around with models. What have you been using this stuff for? What's different now that you have more bandwidth and more time to read and.

[Ben Buchanan]

Just like, yeah, more time, but also, also more access to this stuff. It's crazy that basically for the whole time I was in the White House, this stuff was not accessible on government computers, even on unclassified networks. There's again back to the challenge we were talking about of adoption. And like, we tried to make it a little bit better, but this is a heavy lift So I just have much more time to use this stuff now and I can, I can use this. I love giving, when I write something, I love giving it to Claude and saying, look, you're a really aggressive editor, tell me all the reasons this is wrong or that, that kind of stuff. And I, you know, don't take all of his edits. But I do find that if you tell Claude to be really aggressive, it'll go after your sentence structure. It'll say this is unclear. It'll. It'll say have you thought about this counterpoint? So I really enjoyed just having access to tools like that on a day to day basis. And then I think, you know, I don't do as much coding and the like as I used to, but if I were doing software development, it really does seem like that has just changed everyone's workflow. And there's probably a broader technology lesson from that too.

[Jordan Schneider]

Okay, so you're writing this book about AI. What are the, what are the parts which feel easier to write? What are the parts which you're still noodling on, which feel harder?

[Ben Buchanan]

I think writing about AI is as a whole harder than I expected because of the very same thing that makes AI so interesting to think about and work on, which is everything is interconnected. You have a technology story that's unfolded over a couple decades but really accelerated in the last decade. That's an algorithm story, that's a data story, but it's also its own computing story and the complexity of the compute supply chain. You have a backward looking story, but then you also have the forward looking story of how is this going to get better and recursive self improvement and so forth. You have the core tech and then you have its application to a bunch of different areas. We talked about cyber and then you have a bunch of geopolitical questions, the United States, China, national security, adoption chip controls, all of that. And then you have a bunch of domestic questions. Are AI companies getting too powerful? Will we have new antitrust and concentration of power issues? What's the trade off between privacy and security in the age of AI? The jobs question, the disinformation question, so forth. So there's just, I love it because it's kind of this hyper object where everything is so connected and if you live with the hyper object, you can kind of get a sense of it just by being to mixed metaphors in that mil. But it's so hard to explain to someone who doesn't and it's so hard to figure out what's the. If I've got this huge hand of cards here and they're all connected. What is the way in which I unfold these cards on the, on the page? And that has been the challenge. That's the challenge in teaching it in the classroom and that's the challenge in writing about it. And it's, it's incredibly frustrating. Anyone who tells you otherwise has not done it because there's no easy way to do it. But it does give me even more appreciation for just the depth and breadth of this subject. And I think there's probably a version of this like this is why it's hard from a policy making perspective because it doesn't fit in jurisdictional boundaries and all the mechanisms we've set up to try to govern our processes kind of break when you have something that is so all encompassing.

[Jordan Schneider]

So, you know, you will have written four books in the time in which I will have written zero. I am a lot of what chinatalk does is kind of live at the frontier of that hyper object, whether it's AI or Li, Chinese politics or what have you, but the bid to write a mainstream, you know, something for a, you know, for a trade press about this, you know, these are different from your, for your older books as well. So like, why, you know, what's the, what was the appeal to you of trying to bring a more kind of holistic thesis statement that can be read by more people than already listen to China talk about this topic?

[Ben Buchanan]

I think there's three reasons and I don't know the honest weighting of which one's the most. The first reason is I do think this subject is incredibly important. And I mean, ChinaTalk is going to reach a lot of people. So I'm not comparing audience sizes, but I do think a book length deep dive treatment into this subject that's accessible to a lot of people has value because it's going to touch on my aspects of their lives and of policy. And I just think in a democracy we all kind of have to engage with the most pressing issues and this is, this is going to be on the list. That's 1, 2 is. I think for myself there's a lot of value in refining my own thinking by trying to get it on the page and structure it in a book. And I think in many cases, again, you can live in the milieu and feel like you understand the milieu, but your own thinking just gets so much sharper when you've got to structure it across 300 pages and say what are all the really important things I'M going to leave out and how do I prioritize this and how do I unfold the different pieces and so forth. So that's been incredibly frustrating, but I hope it pays off, not just for the reader, but also for me. And then I think third is there is something about this where I just get great joy out of explaining it or trying to explain it. And insofar as the promotions I got in the White House and the responsibilities I was given by the end being the White House special advisor for AI, I don't think I got that because I had a deeper knowledge, the deepest knowledge of AI in the world. You could take someone from anthropic who could go much, much deeper into how do we do the reinforcement learning step of reasoning models and so forth. I do think my comparative advantage was I could understand it enough and then I could explain it to people who don't work in AI, the President, Jake, Bruce Reed and like, who have to manage the entire world, but who know this is important and want the crisp explanation. So I think frankly I've just gotten a lot of joy from doing that. That's why I'm a professor. That's why I was pressed before the White House. And this feels like going back to that kind of work.

[Jordan Schneider]

Okay, well that's very wholesome. I mean, on the first point though, I guess the sort of like that really opens up a whole new line of like, when are the people going to have a say in all of these discussions? Because we went from this being a thing that a handful of nerds in the bay in Washington cared about to people I think care about from the perspective of it's impacting their 401ks but just starting to sort of really reshape workplaces. And I don't know, I mean, my coming back very early to your framework of like, what is going to drive competition, the sort of inevitable question mark Democratic backlash to the societal changes that AI is going to develop, you know, engender seems to me like one of the biggest unknowns from a US China perspective of like, you're going to have to go through some real social weirdness and real economic dislocation in order to like get the most out of what this technology is going to bring. So yeah, I think it's going to.

[Ben Buchanan]

Be a political issue and I think there'll be a lot of dimensions of AI policy that show up in the 2028 presidential race. Jobs being one, data center, infrastructure being another, probably some national security dimensions to it as well. Child safety another really important dimension, not my field, but one that I imagine is going to resonate in 2028. And I think this is the case where you will see a lot of this show up in the political discussion. And I think, you know, I claim no ability to actually influence the political discussion, but insofar as I can help make it a little bit more informed by the technical facts, especially on the national security side, where I have a little more expertise, I think that's a really important thing to do.

[Jordan Schneider]

Yeah, I don't know. It's just like, yeah, there's a world where it's a big mess and there's a world where. Well, I don't know. Yeah, let's talk a little bit about regulation. So the social media era, basically there was none. And we're kind of dealing with the consequences of that, the sort of shockwaves that we're going to end up engendering again, if AI hits seem to be like an order of magnitude or 2 larger than what we saw from, you know, Facebook and, and Twitter, I mean, is there a, like, the government was just shut down for three months? Like, is there a world in which this ends up turning out where, like, the government really has to play a domestic regulatory role and it like, turns out fine.

[Ben Buchanan]

Where the government has to play a domestic regulatory role of AI?

[Jordan Schneider]

I don't know. That's not the actual question. But I guess the question is, okay, social media came and went without any real kind of domestic regulatory action. What are the tripwires where Ben Buchanan wants to see the government like, start to shape the way in this technology developments and sort of like, when do you think the public is going to kind of like. Like what are the tripwires from the, from a public demand perspective?

[Ben Buchanan]

Yeah, I think a core purpose of the government is to manage these kind of tail risks that affect everyone, but maybe no one else has an incentive to address. And in AI, I think that's things like AI biorisks or AI cyber risks as the technology continues to get better. We took steps on that, using the Defense Production act to get the companies to turn over their safety test results, and President Trump has since repealed those. But I would stand by them as kind of robustly good things to do with very low imposition on the companies. One CEO estimated that the total compliance time for our regulation on AI was something like one employee day per year. So I think pretty reasonable, but also had tractable benefits. What where I'm not is saying the government should be in the business of prescribing the kinds of speech outside of a national security context that AI companies can offer their users and saying, you know, you have to have this political view or you've got to have this kind of take when you're asked this question. That just strikes me as not a road we want to go down based on the evidence I've seen so far. So we try to be very clear in what we did, even on the voluntary side of saying we are focused on things like national security risk, on safety risks and the like. And that strikes me as the right place to start. And I would be hesitant before I would go too much further beyond those kind of core tractable risks.

[Jordan Schneider]

Yeah, I mean there's an interesting sort of US China dynamic here where the AI companion context is the one I think we've spent the most time on. China talk about where I can see just a really dark future where we're all best friends and lovers with AIs that have like enormous power over us. And the Chinese system has shown its willingness to like A, ban porn, B, ban video games for kids outside of like 30 minute windows from like 6:30 to 7:00pm and the sort of the, the it'll be interesting to see if like we end up like, we end up having like a new version of a temperance movement or something where there ends up being some big kind of public demand for government controls or even rejection of some of what ends up being on offer in the coming years.

[Ben Buchanan]

Look, that may happen and I'm not even sure we can debate how it would be good or bad. And there's probably some context in which we could say this is good that, you know, for example, AI systems should not be helping teenagers commit suicide. Like this is, this is not a complicated thing morally in my view. There's just a different question of like, should the federal government be the one doing this and what does it look like to have the federal government do it and so forth. And we didn't really go near any of that right now during our time we, we focused on the national security risks where I think we can all pretty quickly agree like you, yes, this is a core federal compensate to make sure AI systems don't build bioweapons. And frankly the government has expertise around bio that the companies don't have. And the companies were the first ones who told us that and they wanted a lot of assistance and so forth and that's why we created things like the AI Safety Institute.

[Jordan Schneider]

Well, that was a punt, but there better be an AI companion chapter in your new book. Ben, I think there is a.

[Ben Buchanan]

I don't have any developed thoughts on AI companions, except that I actually think generally there is a. I do have concerns about the way in which AI will erode core fundamental pillars of the social contract and social relationships and all that kind of stuff. I absolutely have those concerns.

[Jordan Schneider]

Yeah. I mean, it's like right now we're all walking around with AirPods, but the AirPods are playing music and books and.

[Ben Buchanan]

Podcasts or mind play, China talk.

[Jordan Schneider]

But okay, great. But it's still me on the other side of that, right? And this sort of the level of social anime, Amity anime that we're gonna end up with when it's just optimized. Like, whatever's in your AirPods is just, like, perfectly calibrated for scratching, making every neuron in your brain fire. It's just, It's a weird one. But you said you don't have thoughts on this, so we can, we can, we can move on.

[Ben Buchanan]

No, I don't have smart thoughts on it, but I can appreciate the concerns about AI slop. And I think in the end, the AI companies that are trusted are the ones that are going to be, you know, explicitly humanistic in their values and the like. I think that's right. And I think. I think these ultimately are questions that they're not US Government questions to answer, but these are questions that are US Society questions to answer.

[Jordan Schneider]

Sure. All right, let's close on AI parenting. I bought the Amazon Alexa kids the other day. They had some promotion. It was like 20 bucks.

[Ben Buchanan]

Awful.

[Jordan Schneider]

I was so disappointed.

[Ben Buchanan]

What is the Amazon? As someone who's paranoid about always on microphones to begin with, what is.

[Jordan Schneider]

Well, there's a button.

[Ben Buchanan]

Alexa kids to turn the microphone.

[Jordan Schneider]

Microphone off.

[Ben Buchanan]

Okay.

[Jordan Schneider]

But ostensibly. But even then, it's just like you figured it could talk to you in a normal way. And it's just. It's still really dumb. I mean, it's. It's kind of shocking that, like, there are not smart, sort of like smartphone like, like smart friends already for children. I don't know.

[Ben Buchanan]

I think there probably is a lesson there about AI adoption and diffusion within the economy, which is you have a few companies, Google, OpenAI, Anthropic, that are really inventing the kind of frontier tech, but the actual application of that tech to a lot of different challenges and products and all of that is still very, very nascent and very, very jagged and very uneven in where it's good and where it's not good. I don't know what LLM, if any LLM is in the Amazon Alexa. But I think the general trend is one of we are in the very early innings of applying this stuff even as we're kind of like racing through the movie of inventing more and more powerful versions of it.

[Jordan Schneider]

So Ben, I used to ask people for their songs, favorite songs, but like I can't do that now because we keep getting copyright struck when I put songs after it. So we are now generating cut customize suno songs based on it. So I'm gonna do one about like, you know, creating export controls. But what I need you to do is give me the. The musical genre.

[Ben Buchanan]

The musical genre, yeah. Well, I think, I think it has to be jazz because there clearly was some element in which every policy making process is improvisation and you have some sense of where you're going. But I certainly didn't feel like I had a sheet of music I was reading from. Not that I can read sheet music anyway, but I think it has to be jazz. Jordan.

[Jordan Schneider]

Awesome. All right, Ben. Well, thanks so much for being part of Chinatown.

[Ben Buchanan]

Thanks for having me. This was fun.

[Chinatown Rap Artist]

In the Eisenhower interior Where the blueprints lie a mansion his crew watch the foundries run dry they map the supply chain from ASML to TSML SMC cutting off you lithography node by node See no press release, no spotlight just fab design file targeting SMIC Huay across a thousand miles the default of political technologies grew but the chips they're controlling are the ones that make you move Top cover White House get out of jail free ambassadors fuming bout the 5&M decree Bulldozing antibodies can't keep the pace this ain't a waltz daddy it's a way for Chase October 22 Jake Sullivan laid down the command as large of a lid as possible that's a strategic stand they choke the advanced nose Cut off the supply hand that's the semiconductor siege that's the export band. They recruit the ringers who've run the fab floor From Berkeley clean rooms to K's door While the lifers just rust in their government seat these cats work the choke points with a sink of pay to be they leverage the power on photo resistant gas craft silicone sanctions with lithographic class Long nights, no windows just zealots devotion freezing China's road map from 5 and M to 3 M locomotion Chris McGuire blows the Clarence line Tearing chabber comps on those supply chains Find safe cons on the drums keeping that fat beat bamboo cannon's bass holds down the Mosfet they make the names then disappear that's how you work when the note is clear no elevation, no profile, no brass ring Just the quiet knowledge that you kill the foundry swing so avoid the press like a mask of liners precision stay underground at your clean room vision 10-22- they made the semiconductor bleed with evil restrictions photo mask and seed no marching ban no ticker tape parade Just a small team with a quiet blade Cutting through policy with the rally sages the flare leaving Chinese fabs with legacy nodes in midair here's to the zealots who know the real NO ranges who work the interagency's dangerous changes a rare breed fleeting now gone from the scene they're working the silicone sharp and unseen October 22nd Jake Sullivan laid down the command as large of a lead as possible that's the strategic stand they choke the advanced nose cut off the supply hand that's the semiconductor Dr. Siege that's the export band.