Rudd Warns China Firewall Cannot Stop Rogue AI Amid US Arms Race

The Economist

Beijing risks systemic infrastructure collapse by assuming internet firewalls can contain rogue autonomous models, even as zero-sum rivalry blocks an urgent bilateral safety pact with Washington.

AI-assisted summary of Kevin Rudd on the real risk of China and America’s AI race | The Economist — original source. Check the original for full context.

Source published: 2026-10-01

Key ideas

Formal US-China diplomatic channels on AI governance have opened, but Washington and Beijing remain deeply divided on whether frontier models represent an immediate shared emergency. (0:00)
Beijing mistakenly treats AI risks primarily as an informational threat to regime stability, overestimating the ability of state firewalls to contain autonomous agents attacking critical infrastructure. (1:18)
A sharp internal disconnect exists in both nations between alarmed computer scientists who fear uncontained autonomous models and political leaders fixated on zero-sum dominance. (3:48)
US rhetoric advocating technological supremacy—exemplified by an essay from Anthropic CEO Dario Amodei—provoked heavy Chinese backlash, reinforcing Beijing's belief that Western safety frameworks merely disguise attempts to freeze China in second place. (4:51)
While an AI catastrophe akin to the Cuban Missile Crisis may be required to force bilateral cooperation, China's technocratic leadership may eventually heed its own scientists and seek serious risk-reduction mechanisms. (7:09)

Detailed notes

## Background Published on YouTube by The Economist on October 1, 2026, this interview features geopolitics editor David Rennie speaking with Kevin Rudd, former Australian prime minister and president of the Asia Society. The video examines the intensifying AI race between the United States and China, evaluating whether Washington and Beijing can cooperate on frontier AI safety following a high-stakes summit between Donald Trump and Xi Jinping.

## TL;DR Kevin Rudd analyzes the geopolitical and technical fractures undermining bilateral AI cooperation between the United States and China. He warns that Beijing's leadership suffers from a false sense of security by assuming its cyber defenses can contain rogue frontier models, despite mounting alarm among Chinese computer scientists over autonomous agents escaping human control. Simultaneously, negotiations face gridlock because Chinese realists view American calls for AI guardrails as a pretext to preserve US technological hegemony. Rudd argues that while it may take an AI-driven crisis on par with the Cuban Missile Crisis to force meaningful alignment, China's technocratic governance may ultimately recognize the shared existential threat and engage in pragmatic risk reduction.

## Atmosphere & style The conversation carries an urgent, sobering, and intensely analytical tone, characteristic of elite geopolitical discourse dissecting an existential threat. David Rennie's incisive, occasionally dry questioning meshes seamlessly with Kevin Rudd's authoritative, insider fluency in Chinese political psychology. The dialogue balances cold strategic realism with creeping dread, conveying the unsettling disconnect between high-level diplomatic posturing and the uncontrollable velocity of frontier artificial intelligence.

## Bulletin - [00:00](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=0s) Formal US-China diplomatic channels on AI governance have opened, but Washington and Beijing remain deeply divided on whether frontier models represent an immediate shared emergency. - [01:18](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=78s) Beijing mistakenly treats AI risks primarily as an informational threat to regime stability, overestimating the ability of state firewalls to contain autonomous agents attacking critical infrastructure. - [03:48](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=228s) A sharp internal disconnect exists in both nations between alarmed computer scientists who fear uncontained autonomous models and political leaders fixated on zero-sum dominance. - [04:51](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=291s) US rhetoric advocating technological supremacy—exemplified by an essay from Anthropic CEO Dario Amodei—provoked heavy Chinese backlash, reinforcing Beijing's belief that Western safety frameworks merely disguise attempts to freeze China in second place. - [07:09](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=429s) While an AI catastrophe akin to the Cuban Missile Crisis may be required to force bilateral cooperation, China's technocratic leadership may eventually heed its own scientists and seek serious risk-reduction mechanisms.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=0s) The United States and China dominate an unprecedented technological revolution where leading executives estimate a 10% probability of human extinction. - [00:00](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=0s) High-level bilateral dialogue opened following an initial May agreement, though Donald Trump signaled comfort with the status quo while Xi Jinping framed himself as advocating human control. - [01:18](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=78s) Chinese political leaders view AI primarily through the lens of regime security and unauthorized information diffusion rather than systemic technological collapse. - [01:18](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=78s) Beijing's reliance on cyber firewalls is a dangerous miscalculation, as existing defenses cannot protect critical economic, social, or military infrastructure from out-of-control frontier models. - [03:48](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=228s) Neither Washington nor Beijing has established an internal political consensus regarding how to govern existential AI risks. - [03:48](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=228s) Chinese computer scientists have become acutely alarmed that rapid breakthroughs could spawn autonomous agents capable of mutating and escaping human oversight. - [04:51](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=291s) Both powers share a realist operational assumption that the AI race is entirely zero-sum and that whoever dominates the technology stack will dictate global power. - [04:51](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=291s) Paralleling historical nuclear arms control, Beijing rejects negotiating constraints from a position of technological inferiority to avoid locking in US supremacy. - [04:51](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=291s) Western arguments for bilateral guardrails coupled with calls to maintain a perpetual US technological lead sparked intense backlash across China's policy ecosystem. - [07:09](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=429s) A severe real-world failure or an AI equivalent of the Cuban Missile Crisis may prove necessary to shock both superpowers into prioritizing shared survival over strategic competition. - [07:09](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=429s) Kevin Rudd maintains cautious optimism that China's technocratic governance will eventually recognize the physical reality of frontier model risks and engage in substantive risk-reduction talks.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=0s) Bilateral AI Talks and Beijing's False Sense of Security

David Rennie frames the discussion around an extraordinary historical juncture: two superpowers, the United States and China, utterly dominate a technology so disruptive that leading tech executives warn of a 10% chance of human extinction. Yet when Donald Trump and Xi Jinping recently met—likened by Rennie to "two kings having this kind of parley with their knights"—they appeared to be speaking past each other. Trump signaled satisfaction with the status quo, while Xi emphasized maintaining human control over artificial intelligence. Rennie questions whether Xi’s remarks reflected genuine strategic concern or merely posturing to seize the moral high ground as the more responsible leader.

Kevin Rudd responds that while it is too early to know definitively, the crucial development is that a formal diplomatic channel has opened. A headline agreement reached in May was subsequently operationalized. However, the core question remains whether Washington and Beijing share any real consensus on the level of emergency posed by frontier models—whether developed through proprietary frameworks in the US or open-source, open-weights models in China. Rudd points out a dangerous flaw in Beijing's thinking: China's leaders have long treated AI primarily as a regime security issue—namely, unauthorized information dissemination within a one-party state. Believing themselves to be "masters of the firewall," Chinese officials assume they can simply turbocharge their existing cyber defenses against rogue autonomous agents or malicious state actors. Rudd considers this assumption a grave miscalculation:

> "I think that's a Chinese error because even their cyber defenses, in my understanding, are not sufficiently robust to deal with the wholesale assault on social infrastructure, economic infrastructure, even military infrastructure, which could be brought about under given circumstances by some of the most advanced frontier models at present were they to escape human control."

[03:48](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=228s) The Technologist-Policymaker Divide

When asked whether both sides are simply retreating into familiar political comfort zones rather than confronting systemic catastrophe, Rudd highlights an internal disconnect within both nations. Neither the American nor the Chinese system has reached an internal policy consensus. Instead, a clear mismatch exists between the technologists who build frontier models and the political leadership tasked with governing them.

In China, one does not have to venture far outside the leadership compound of Zhongnanhai toward the tech sector to encounter acute alarm. Between the May meeting in Beijing and the September summit in Washington, the rapid pace and intensity of technical breakthroughs have thoroughly spooked Chinese computer scientists. Their primary dread is that frontier models will generate autonomous agents capable of independent action. Once diffused and adapted globally—or even within China—these agents could mutate beyond human oversight and turn back against the Chinese state. Within the political elite, however, the reflex remains anchored in political self-preservation. Accustomed to treating informational containment as an American problem unique to open societies, Chinese policymakers are only beginning to grapple with the reality that existential AI risks cannot be stopped by internet filters.

[04:51](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=291s) The Zero-Sum Arms Race and the Dario Amodei Dilemma

Rennie turns to the fundamental diplomatic paradox: can substantive negotiations occur while the Trump administration openly insists on preserving American technological supremacy? In nuclear arms control, Beijing has historically refused to negotiate from a position of numerical inferiority, operating on the conviction that the weaker party in any negotiation will simply have an unfavorable deal imposed upon it. With American leaders and prominent figures like Anthropic CEO Dario Amodei arguing that the US must keep China in second place, Rennie asks if fear alone can push China to the table.

```mermaid flowchart TD accTitle: The AI Arms Race Trap accDescr: Diagram showing how technological rivalry and calls for guardrails create mutual distrust between the US and China.

subgraph US["United States Posture"] US_Lead["Pursue Absolute AI Dominance Across the Stack"] US_Guard["Call for Bilateral Guardrails & Control"] end

subgraph CN["Chinese Strategic Calculus"] CN_Realism["View AI as Zero-Sum: 'Whoever Wins AI Wins'"] CN_Parity["Refusal to Negotiate from Weakness or Accept Constraints"] end

US_Lead -->|"Reinforced by Calls to Keep China Behind"| CN_Parity US_Guard -->|"Perceived as Pretext to Freeze US Advantage"| CN_Realism CN_Realism --> Friction["Mutual Distrust & Diplomatic Gridlock"] CN_Parity --> Friction ```

Rudd confirms that China's default operational reality mirrors Trump's view: both sides view the situation as an all-out strategic competition across the entire technology stack where "whoever wins AI wins." While Chinese rhetoric may sound more measured, realists recognize that Beijing has no intention of being bound by external constraints. When American voices advocate for bilateral guardrails while simultaneously publishing essays—such as Amodei's recent piece arguing that the strategic goal must include staying ahead of Beijing—Chinese leaders interpret Western appeals for safety as a thinly disguised effort to lock in American hegemony.

- Rudd notes that the Chinese system erupted with an intense, voluminous backlash against Amodei's essay. - Rennie quips that bipartisan Washington and Beijing have found one rare point of agreement: "They all hate Dario Amodei." - Despite personal chemistry between Trump and Xi, strategic distrust makes both governments unwilling to tie their own hands while playing catch-up.

[07:09](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=429s) Catalysts for Governance: Catastrophe vs. Scientific Rationalism

Rudd poses the existential dilemma facing both powers: will it take a catastrophic real-world disaster to finally force Washington and Beijing into collective action? He asks whether humanity must endure an outright failure or an AI-driven equivalent of the Cuban Missile Crisis—an event that brings civilization to the brink of annihilation—before the two rivals experience a shared "come-to-Jesus, come-to-Marx moment" and admit that the threat is bigger and uglier than their bilateral rivalry.

Alternatively, Rudd asks whether the two sides can act like "grown-ups" by listening to their technologists and preemptively mitigating the risk before an uncontrollable failure occurs. Despite the obstacles, Rudd expresses cautious optimism. Because China possesses an intensely science-based political culture and a technocratic leadership cadre, he believes Chinese officials will soon conclude that the threat of frontier autonomous models is urgent, material, and real. Once that internal threshold is crossed, Beijing will be motivated to operationalize concrete negotiating mechanisms with the Trump administration, paving the way for serious bilateral risk-reduction talks.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=0s) two kings having this kind of parley with their knights — David Rennie

> [00:00](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=0s) I think that's a Chinese error because even their cyber defenses, in my understanding, are not sufficiently robust to deal with the wholesale assault on social infrastructure, economic infrastructure, even military infrastructure, which could be brought about under given circumstances by some of the most advanced frontier models at present were they to escape human control. — Kevin Rudd

> [00:00](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=0s) masters of the firewall — Kevin Rudd

> [04:51](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=291s) whoever wins AI wins — Kevin Rudd

> [04:51](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=291s) They all hate Dario Amodei. — David Rennie

> [07:09](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=429s) come-to-Jesus, come-to-Marx moment — Kevin Rudd

> [07:09](https://www.youtube.com/watch?v=a4wrmRXyxlE&t=429s) grown-ups — Kevin Rudd

## Mentioned Kevin Rudd, David Rennie, Donald Trump, Xi Jinping, Dario Amodei, Anthropic, Asia Society, The Economist, Zhongnanhai

## Video description America and China are locked in a race for AI dominance. As the warnings about an AI catastrophe get louder, whether the two countries can work together is becoming an increasingly pressing issue.

Few people are better placed to discuss this subject than Kevin Rudd. Australia’s former prime minister, who is now president of Asia Society, has spent decades studying how Xi Jinping thinks. As Australia’s ambassador to Washington, Mr Rudd also watched Donald Trump’s White House at close quarters. He joins David Rennie, The Economist’s geopolitics editor, to assess the recent summit between Xi Jinping and Donald Trump and examine what it means for the world’s most consequential relationship.

#America #China #AI #politics #technology

00:00 - Can China and America collaborate on AI? 01:18 - How Beijing and Washington see the AI extinction risk 03:48 - The gap between technologists and policymakers 04:51 - Can talks succeed while both sides want to win the race? 07:09 - Will it take an AI disaster for the US and China to talk?

Watch the full show: http://bit.ly/4B0tKCP Sign up to the Insider newsletter: https://econ.st/4nOyzIb

Subscribe to The Economist: https://www.economist.com/subscribe Download our app: https://www.economist.com/get-the-app

Follow The Economist on Instagram: https://www.instagram.com/theeconomist/ Follow The Economist on Facebook: https://www.facebook.com/TheEconomist Follow The Economist on X: https://x.com/TheEconomist

[Source](https://www.youtube.com/watch?v=a4wrmRXyxlE)

中文简报

陆克文警告：美中军备竞赛之际，中国防火墙无法阻挡失控AI 北京假定互联网防火墙能够遏制失控的自主模型，这可能带来系统性基础设施崩溃的风险，即便零和竞争正阻碍其与华盛顿达成紧迫的双边安全协议。陆克文（Kevin Rudd）分析了破坏美中双边人工智能合作的地缘政治与技术分歧。他警告称，尽管中国计算机科学家对自主智能体脱离人类控制的担忧与日俱增，但北京领导层假定其网络防御能够遏制失控的前沿模型，从而陷入了虚假的安全感。与此同时，由于中国现实主义者将美国对AI护栏的呼吁视为维护美国技术霸权的借口，谈判陷入僵局。陆克文认为，虽然可能需要一场堪比古巴导弹危机的AI危机才能迫使双方形成实质性协调，但中国的技术官僚治理模式最终可能会认识到这一共同的生存威胁，并参与务实的风险降低合作。美中关于AI治理的正式外交渠道已经开启，但华盛顿和北京在前沿模型是否构成迫在眉睫的共同紧急威胁上仍存在严重分歧。 (0:00) 北京错误地主要将AI风险视为对政权稳定的信息威胁，高估了国家防火墙遏制攻击关键基础设施的自主智能体的能力。 (1:18) 两国国内均存在严重脱节：计算机科学家对无法遏制的自主模型深感担忧，而政治领导人则执迷于零和主导权。 (3:48) 美国鼓吹技术霸权的言论——例如Anthropic首席执行官达里奥·阿莫代伊（Dario Amodei）的一篇文章——引发了中国的强烈反对，加深了北京的看法，即西方安全框架只是为了将中国冻结在从属地位而作的伪装。 (4:51) 尽管可能需要一场类似于古巴导弹危机的AI灾难才能迫使双边展开合作，但中国的技术官僚领导层最终可能会听取本国科学家的意见，并寻求严肃的风险规避机制。 (7:09)

Reading summary

According to Kevin Rudd, the United States and China are locked in an all-out, zero-sum race for frontier AI dominance, hindered by Beijing's flawed belief that its domestic firewall can contain rogue systems and deep suspicion that American safety proposals are a pretext to preserve technological hegemony. Beijing's Misplaced Reliance on Cyber Defenses (0:00) Although a formal diplomatic channel opened following an agreement in May, Kevin Rudd argues that bilateral engagement is undermined by a fundamental flaw in Beijing’s strategic thinking. Chinese leadership has historically approached artificial intelligence as an internal regime security problem focused on controlling unauthorized information flows within a one-party state. Viewing themselves as 'masters of the firewall,' Chinese officials assume they can neutralize threats from autonomous agents or hostile state actors simply by strengthening existing cyber defenses. Rudd characterizes this assumption as a grave miscalculation. In his assessment, China's cyber defenses are inadequate to protect social, economic, and military infrastructure from the systemic disruption that advanced frontier models—whether proprietary US frameworks or Chinese open-weights systems—could inflict if they escape human oversight. The Divide Between Technologists and Political Leadership (3:48) Rudd observes that neither Washington nor Beijing has established an internal policy consensus on AI governance, resulting in a pronounced disconnect between the technologists developing frontier models and the politicians tasked with regulating them. In China, technical breakthroughs between the bilateral meetings in May and September generated severe alarm among computer scientists outside the leadership compound of Zhongnanhai. These researchers fear that frontier models will produce autonomous agents capable of independent mutation and action, potentially turning against the state. Conversely, China's political elite remains focused on political self-preservation. Because Chinese officials have long viewed informational containment as a vulnerability specific to open Western societies, they have been slow to realize that the existential risks posed by uncontrollable frontier models cannot be mitigated through standard internet filtering mechanisms. Strategic Distrust and the AI Parity Dilemma (4:51) Substantive negotiations face a diplomatic paradox rooted in historical precedent: much like its posture in nuclear arms control, Beijing refuses to negotiate from a perceived position of technological weakness out of concern that an unfavorable settlement will be imposed. David Rennie and Rudd note that both the Trump administration and Chinese leadership operate on the realist assumption that 'whoever wins AI wins' across the entire technology stack. This zero-sum posture fuels mutual distrust. When American industry figures like Anthropic CEO Dario Amodei publish essays advocating that the United States must maintain its technological lead over China, Beijing interprets Western proposals for bilateral guardrails as a disguised strategy to lock in permanent American hegemony. Rudd notes this dynamic sparked an intense backlash within China, leaving both sides reluctant to accept operational constraints while actively competing for superiority. Catastrophic Catalysts versus Technocratic Rationalism (7:09) Rudd outlines two potential pathways toward meaningful bilateral AI governance: a reactive response triggered by an acute, systemic disaster, or a proactive shift driven by scientific consensus. He questions whether the two superpowers will require an AI-driven crisis analogous to the Cuban Missile Crisis—bringing civilization to the brink of catastrophe—before finding the political will to acknowledge that existential technological risks transcend their geopolitical rivalry. Despite these tensions, Rudd offers cautious optimism for the proactive path. He argues that China’s technocratic leadership and science-oriented political culture will eventually force officials to recognize the material risks posed by autonomous frontier models. Rudd anticipates that once this internal threshold is crossed, Beijing will be motivated to establish functional negotiating mechanisms with the Trump administration to reduce bilateral risks. While high-level diplomatic channels between the United States and China have opened, genuine risk mitigation remains stalled by zero-sum competition and mutual suspicion. The fundamental unresolved question is whether technocratic rationality will lead both powers to implement bilateral guardrails preemptively, or whether substantive cooperation will emerge only in the aftermath of an uncontrolled technological catastrophe.

Source quotes

[0:00] David Rennie: two kings having this kind of parley with their knights [0:00] Kevin Rudd: I think that's a Chinese error because even their cyber defenses, in my understanding, are not sufficiently robust to deal with the wholesale assault on social infrastructure, economic infrastructure, even military infrastructure, which could be brought about under given circumstances by some of the most advanced frontier models at present were they to escape human control. [0:00] Kevin Rudd: masters of the firewall [4:51] Kevin Rudd: whoever wins AI wins [4:51] David Rennie: They all hate Dario Amodei. [7:09] Kevin Rudd: come-to-Jesus, come-to-Marx moment [7:09] Kevin Rudd: grown-ups

Original transcript

[0:00] You have this extraordinary moment where [0:01] you have this potentially [0:03] extraordinarily disruptive technology [0:05] and two countries utterly dominant, [0:08] America and China. And you know, we've [0:10] had all these tech bosses saying that [0:12] you know, there might be a 10% chance of [0:13] the extinction of humanity. [0:15] And yet these kind of these two kings [0:17] having this kind of parley with their [0:18] knights, they didn't seem to be [0:20] particularly engaging at the same level. [0:21] You know, Trump was talking about [0:23] essentially we we like things as they [0:25] are. Xi Jinping said some language [0:26] around you need to have human control of [0:29] AI. Was that signaling a completely [0:31] different view or was that just him kind [0:33] of taking the moral high ground, [0:35] positioning himself as a bit more [0:36] responsible than Donald Trump about [0:38] this? [0:39] >> I think it's uh generally too early to [0:41] tell. What I am happy about is that this [0:44] channel has been opened. Uh there was a [0:46] headline agreement to do so back in May. [0:50] Now the channel, as I understand it, is [0:52] being operationalized. [0:54] The more important and fundamental [0:56] question is is there a consensus about [0:59] the level of emergency which uh frontier [1:02] models now represent, those both being [1:05] generated through proprietary models in [1:08] the US or through open source open [1:10] weights models in the case of China. I [1:13] think because China has internally [1:15] concluded that its own cyber defenses [1:17] are already robust and that that in turn [1:21] comes from their view that AI has always [1:23] represented a problem in terms of [1:26] unauthorized information dissemination [1:28] within a one-party state and therefore [1:31] we are the masters of the firewall, you [1:33] ain't seen nothing yet until we've [1:34] further turbocharged our firewall around [1:37] cyber defenses against anything [1:39] malevolent which could come out of uh [1:42] the autonomous uh uh actions of AI [1:45] agents, let alone if they're being [1:47] manipulated by third parties, rogue [1:49] actors, rogue states. [1:51] I think that's a Chinese [1:53] uh error because even their cyber [1:56] defenses uh in my understanding, are not [1:59] sufficiently robust to deal with the [2:02] wholesale assault on social [2:04] infrastructure, economic infrastructure, [2:05] even military infrastructure, which [2:07] could be brought about under given [2:09] circumstances by some of the most [2:12] advanced frontier models at present were [2:15] they to escape human control. [2:17] >> Do you worry that each side is [2:18] retreating to its kind of comfort zone [2:20] and actually there's something much much [2:21] larger at stake and neither side is for [2:23] the moment wrestling with that? [2:26] >> My view is that neither the Chinese [2:27] system nor the US system has reached [2:30] final consensus internally on this yet. [2:34] I think there is something of a mismatch [2:36] between the technology community in both [2:39] countries on the one hand and the policy [2:42] community and political community on the [2:44] other. [2:45] In the case of China, you don't have to [2:47] go very far out of Zhongnanhai and [2:50] heading towards [2:52] the Chinese technology world to reach a [2:54] conclusion that this is potentially [2:58] really quite dangerous. And in fact, the [3:01] developments which have unfolded in the [3:03] technology itself between the May summit [3:06] in Beijing and this September summit in [3:09] Washington have been of a pace and [3:12] intensity which have raised real [3:15] concerns in the Chinese technical [3:16] community. [3:18] Can the policy and political [3:19] >> What are they concerned about? [3:20] >> They're concerned precisely about the [3:22] fact that certain frontier models [3:25] produce agents which can act [3:26] autonomously and that frankly, given the [3:30] dissemination of these agents either [3:33] through the US system or even the [3:35] Chinese system elsewhere in the world, [3:37] that they could be further developed [3:40] through diffusion and adaptation in a [3:43] manner which could be used back against [3:45] China itself. [3:46] >> But you're a man who uses words very [3:47] precisely and you said that that concern [3:49] you had heard from the kind of the big [3:51] Chinese AI companies, but when you met [3:54] senior officials, are they more are they [3:57] focused on that to the to the to a [3:59] degree that makes you comfortable or are [4:00] you a bit anxious that they're [4:02] they're still thinking in terms of kind [4:04] of political security? [4:05] >> As I was saying before, I don't think [4:07] they've reached consensus in the same [4:09] way as I see this in the US system [4:11] between the technologists on the one [4:13] hand [4:14] and the policy establishment on the [4:16] other [4:17] to wrap your head around how real is the [4:20] threat, how urgent is the threat, and [4:22] then what to do about it. [4:24] I see those [4:27] tensions in the Chinese system as we [4:29] speak, and to some extent they're [4:30] natural because this is so rapidly [4:33] unfolding because as you rightly said [4:35] before, David, this is challenging the [4:38] historical view that we've got one thing [4:41] to be worried about, it's regime [4:42] security, unauthorized information [4:44] dissemination, we've locked that down, [4:46] thank you very much. This is an American [4:48] open society, open political system [4:50] problem. [4:51] >> Is there any substantive hope of these [4:53] two countries having urgent talks [4:55] as long as the same time President Trump [4:57] and his staff talk about the need to [5:00] maintain American dominance? Because in [5:02] any other sphere you can think about, [5:04] when you're trying to get the Chinese to [5:05] engage on say nuclear arms control, the [5:07] very strong message is [5:09] is as long as we don't have parity, as [5:11] long as you have 10 times more warheads [5:12] than we do, why would we even begin that [5:15] negotiation? Because at the deepest [5:17] level the Chinese belief is that the [5:19] weaker party that enters the negotiation [5:21] will have a bad deal imposed on them. So [5:23] as long as China is the catching up AI [5:26] power and the Americans are openly [5:28] talking, whether it's Trump or actually [5:29] Dario Amodei, about the need to keep the [5:31] Chinese in second place, [5:33] is that a kind of [5:35] total obstacle to really substantive [5:37] talks or you could see them getting [5:39] frightened enough that they'll talk? [5:41] >> The default Chinese position is that is [5:44] exactly the the as President Trump's [5:46] position. This is an active strategic [5:48] competition in AI across the stack, and [5:51] whoever wins AI wins. And whatever the [5:55] polite modulations of the Trumpian [5:57] expression the Chinese may use, that's [5:59] their operational position. Uh anyone [6:02] who's a realist and looking at the uh [6:04] behavior of the Chinese system will [6:06] reach that conclusion. Secondly, you're [6:09] absolutely right to say that when they [6:11] hear the Americans say, "Let's have some [6:13] guardrails. Let's have a bilateral uh [6:17] negotiated framework." And then when [6:19] that's reinforced with uh Dario Amodei's [6:22] uh most recent article saying that the [6:25] strategic objective here is in part not [6:27] just to protect ourselves, but to remain [6:29] ahead of the Chinese, the Chinese are [6:31] not dumb, they read that and they say, [6:32] "Haha. Um that is exactly what we [6:35] thought. Hence the voluminous reaction [6:38] from the Chinese system to Dario's [6:40] proposal, almost as intense as President [6:43] Trump's." [6:43] >> Yeah. Which is the one thing they agree [6:44] on. [6:45] >> They all hate They all hate Dario [6:46] Amodei. [6:46] >> And so therefore [6:48] uh it is very much the Chinese default [6:50] position. Not only are we in a [6:52] competitive relationship, but we're not [6:54] about to have our hands tied on this. [6:57] And so uh here is the dilemma we all [6:59] face therefore, given the mutual [7:02] distrust between both systems, [7:04] notwithstanding the personal chemistry [7:06] apparently between the two leaders. [7:09] Uh do we have to have um a complete and [7:12] utter disaster in the real world to [7:15] provide the galvanizing moment [7:17] um [7:18] to cause Beijing and Washington to have [7:22] a come-to-Jesus, come-to-Marx moment on [7:25] this uh and say, "My God, this is bigger [7:28] and uglier than both of us, and we need [7:30] rapidly to collaborate." Or short of [7:33] that, are we saying we've got to have a [7:35] Cuban missile crisis which almost [7:37] eliminates humanity um in order to bring [7:40] us to the same come-to-Jesus, come to [7:42] Marx moment, or can we be a little more [7:44] grown up than that [7:46] and actually realize from what the [7:48] technologists say say to us that this is [7:51] a real world prospect and it's far [7:52] better we get ahead of the pace now. [7:55] I'm optimistic that given China's [7:58] intense science-based culture and [8:01] technology heavy leadership that they'll [8:03] reach a conclusion soon that this is [8:05] material and real and urgent that [8:08] they'll be able to operationalize [8:11] the negotiating track with President [8:13] Trump, just as I'm optimistic that the [8:15] US side of this argument can similarly [8:18] engage.