Two people in a cave with a laptop could disable any country's power grid
Más de medio siglo después de que Nixon volara a Pekín para romper décadas de hostilidad, Trump y Xi se preparan para un encuentro que podría definir una nueva era de cooperación forzada. La amenaza no es una nación rival, sino una tecnología —la inteligencia artificial con capacidades autónomas— que puede ser empuñada por actores pequeños para paralizar infraestructuras críticas de cualquier país, sin recursos significativos y sin temor a represalias. Como ocurrió con la disuasión nuclear durante la Guerra Fría, Estados Unidos y China comparten ahora una vulnerabilidad común que ninguno puede gestionar en solitario, y el tiempo para actuar antes de que el control se vuelva imposible se agota con rapidez.
- Sistemas de IA desarrollados por empresas como Anthropic y OpenAI ya pueden detectar y explotar vulnerabilidades de software con una precisión alarmante, poniendo armas cibernéticas devastadoras al alcance de cualquier actor con una laptop y conexión a internet.
- El equilibrio de destrucción mutua que mantuvo estable la guerra cibernética entre superpotencias se derrumba ante grupos que no tienen nada que perder y, por tanto, no pueden ser disuadidos por ninguna amenaza de represalia.
- La proliferación es inminente: una vez que estos modelos se distribuyan ampliamente, la contención será imposible, y el momento en que los gobiernos aún pueden establecer reglas se cierra con cada semana que pasa.
- El encuentro Trump-Xi en Pekín representa una oportunidad histórica para construir marcos conjuntos de gobernanza de IA, involucrando tanto a gobiernos como a las grandes corporaciones tecnológicas de ambos países.
- Sin un acuerdo concreto —no solo una mención en la agenda, sino una decisión de colaborar de inmediato— el mundo podría perder la única ventana disponible para evitar que las armas cibernéticas autónomas queden fuera de todo control.
La próxima semana, Donald Trump se sentará con Xi Jinping en Pekín en lo que podría ser el encuentro más trascendente entre líderes estadounidenses y chinos desde que Nixon voló a esa misma ciudad en 1972 para reunirse con Mao. Aquel histórico giro creó una alianza tácita contra la Unión Soviética. Hoy, más de cincuenta años después, ambos líderes enfrentan una simetría diferente: no un rival estatal, sino una tecnología que ninguno puede controlar por sí solo.
La amenaza es la inteligencia artificial con capacidades autónomas. Sistemas recientes han demostrado que pueden detectar y explotar vulnerabilidades de software con precisión inquietante. En manos equivocadas —terroristas, redes criminales, estados marginales— estas herramientas podrían convertirse en armas capaces de inhabilitar la red eléctrica de cualquier país. La barrera de entrada se ha derrumbado: el conocimiento técnico requerido se evapora, el costo es insignificante.
Durante décadas, Estados Unidos y China libraron una guerra cibernética en las sombras, pero ambos respetaban una lógica de disuasión mutua. Ese equilibrio está a punto de romperse. El nuevo peligro proviene de actores que no están sujetos a esa lógica: grupos pequeños, células disidentes, naciones menores con ambiciones. Ellos pueden empuñar armas de IA sin temor a represalias porque no tienen nada que perder.
La ironía es que las empresas estadounidenses y chinas fueron las primeras en construir estos sistemas. Anthropic y OpenAI han restringido por ahora la distribución de sus modelos más poderosos, pero esa ventana se cierra. Una vez que estas herramientas proliferen, la contención será imposible. Por eso, Craig Mundie —ex director de investigación de Microsoft— argumenta que Trump y Xi deben actuar juntos ahora. No porque la cooperación sea fácil, sino porque la alternativa es peor.
Lo que haría verdaderamente histórica esta cumbre no es que el tema aparezca en la agenda, sino que se tome la decisión de colaborar de inmediato: gobiernos y corporaciones tecnológicas de ambos países moviéndose al unísono. Vivimos en la primera era en que los destinos de las naciones están genuinamente entrelazados. O construimos coaliciones para navegar juntos este momento, o fracasamos juntos. Esa elección se está tomando ahora.
Next week, Donald Trump will sit down with Xi Jinping in Beijing for what could be the most consequential meeting between American and Chinese leaders since Richard Nixon flew to the same city in 1972 to meet Mao Zedong. That earlier summit broke open decades of hostility and created an unspoken alliance: two superpowers united against a common adversary, the Soviet Union. Now, more than fifty years later, Trump and Xi face a different kind of symmetry. They have a new shared threat—not a rival nation, but a technology that neither can control alone.
The threat is artificial intelligence with autonomous capabilities. Recent systems developed by Anthropic and OpenAI have demonstrated something alarming: they can detect and exploit software vulnerabilities with stunning precision. In the hands of the wrong actors—terrorists, anarchists, criminal networks, rogue states—these tools could become weapons. Two people in a cave with a laptop, access to the latest AI models, and a Starlink terminal could theoretically disable the power grid of any country on Earth. The barrier to entry is collapsing. The technical knowledge required is evaporating. The cost is negligible.
For decades, the United States and China have engaged in a shadow war of cyberattacks and espionage. They have probed each other's networks, planted malware, stolen secrets. But both sides understood the rules of this game: if China darkened American cities, America could darken Beijing. It was a form of mutually assured destruction, borrowed from the nuclear age. The calculus held. The balance, however fragile, remained stable. That equilibrium is about to shatter.
The new danger comes from actors who are not bound by the logic of deterrence. A small cell of hackers, a dissident group, a minor nation with ambitions—they could wield AI-powered cyber weapons without fear of retaliation because they have nothing to lose. They exist outside the framework that has kept the superpowers in check. Craig Mundie, the former head of research and strategy at Microsoft, frames it plainly: the United States and China have recreated the conditions of Cold War nuclear standoff, but now a third party is crashing the dinner. And this third party could threaten both of them simultaneously.
The irony is that American and Chinese companies have been the first to build these systems. Anthropic and OpenAI have chosen to restrict distribution of their most powerful models for now, but that window is closing. Other American systems—Google's Gemini—will soon offer the same capabilities. Chinese AI models are not far behind. Once these tools proliferate, containment becomes impossible. The knowledge will spread. The code will leak. The moment of control will pass.
Mundie argues that this is precisely why Trump and Xi must act now, and together. Not because cooperation between Washington and Beijing is easy—it is not. But because the alternative is worse. During the Nixon era, the two countries found common cause against the Soviet Union. Today, they must find it against a technology that respects no borders and recognizes no sovereignty. The old G2—America and China—needs to work with what Mundie calls the new I7: Anthropic, Google, OpenAI, Meta, Alibaba, DeepSeek, and ByteDance. Governments cannot solve this alone. Neither can corporations. Both must move together, and quickly.
There are signs this might happen. Trump is reportedly considering imposing oversight on AI models before they are released to the public—a sensible step. The expectation is that autonomous AI threats will appear on the agenda when Trump and Xi meet. But appearing on an agenda is not enough. What would make this summit truly historic is not merely discussion, but a decision to collaborate immediately. The window for action is closing faster than most leaders in Washington, Beijing, or Moscow seem to understand. We live in the first era of human history when our fates are genuinely intertwined—when no nation, however powerful, can solve the great challenges alone. Either we build coalitions to navigate this together, or we fail together. The choice is being made now.
Citações Notáveis
If China darkened American cities, America could darken Beijing. But now a third party is crashing the dinner—and they could threaten both simultaneously.— Craig Mundie, former head of research and strategy at Microsoft
This is the first era of human history in which we must govern, innovate, and collaborate at planetary scale to survive. Either we build coalitions to navigate this together, or we fail together.— Analysis in the piece
A Conversa do Hearth Outra perspectiva sobre a história
Why does this feel different from other cybersecurity warnings we've heard before?
Because the barrier has collapsed. For years, launching a sophisticated cyberattack required state resources, technical expertise, years of planning. Now an AI system can do the reconnaissance and exploitation for you. A person with no hacking background could potentially cause massive damage.
But couldn't the same AI systems be used to defend against attacks?
Yes, and that's part of why both countries need to move together. If they're racing separately, they both lose. If they coordinate on safeguards and detection, they both gain. Right now they're still in the old Cold War mindset of mutual suspicion.
What happens if they don't cooperate?
The technology spreads anyway. Criminal groups, smaller nations, extremist cells—they all get access. Then you have a world where any actor with modest resources can threaten critical infrastructure. Neither superpower can protect itself alone.
Is this really comparable to the Nixon-Mao moment?
In one way, yes. Both moments involve two rivals recognizing a threat bigger than their rivalry. But this threat isn't another country. It's a tool that could be weaponized by anyone. That's actually more destabilizing.
What would real cooperation look like?
Joint oversight frameworks. Shared intelligence on where these systems are spreading. Agreements on what capabilities get restricted and how. It sounds simple, but it requires trust neither country naturally has.
And if they wait?
Then the decision gets made for them. Once these tools are in the wild, you can't put them back. The moment to shape what happens is now.